Skip to content

Duplicate detection functions

Module containing functions for finding and removing the duplicates among the input databases. Used for e.g. dup_detection.ipynb, dup_decision.ipynb, dup_removal.ipynb)

This script includes functions which search for duplicates.

Last updated 19/12/2025 by LL for publication of dod2k v2.0

Update 22/10/24 updated duplicate_decisions: - created backup decision file which is intermediately saved - outputs URL which can be copied and pasted into browser - implemented a composite option in the decision process, to create a composite of two records - removed date (YY-MM-DD) of decision output filename Update 8/10/24 changed colours of figure in dup_plot (dropped highlighting differing metadata). replaced db_name with df.name

Update 2/10/24 Implemented a commenting option in duplicate_decisions to comment on decision process and on individual decisions.

Update 27/9/24 Updated directory names and changed the correlation and distances output in find_duplicates to only output data from potential duplicates (replaced saving all pairs)

Update 9/9/24 Introduced timestamps and contact details into duplicate decision output csv and changed the file= and dirnames for streamlining purposes. Update 23/8/24 Replaced the function cleanup_database and split into two: plot_duplicates: plots the candidate pairs, saves figures and a summary csv sheet. cleanup_database_2: goes through the candidate pairs and makes decisions based on the options: a) raw input b) keep all records c) automatically keep only updated records and eliminate the other candidate. Decisions and metadata are saved in csv file.

Update 22/8/24: Fixed a bug in find_duplicates logical algorithm- wrong bracket closure. Also changed location_crit to account for nans in elevation. Calculation of z-scores only divided by std if std!=0 to avoid nans.

Update 15/8/24: updated numerical checks for duplicate detection: implemented check for correlation and rmse of records plus correlation and rmse of first difference.

Update 13/8/24: updated keys to account for updated dataframe terminology. Revised loading/saving of data in find_duplicates find_duplicates: updated the logic for duplicate detection: overlap_crit now accounts for short records (allows short records to pass through without overlap check) corr_crit only one numerical criterion needs to be satisfied for the data (either correlation or rmse or 1st difference) location_crit now includes elevation too

cleanup_database: updated plot to include URL, Database and streamlined table Script written by Lucie Luecke, 6/7/23

collect_dup_details(df_decisions, header)

Generate duplicate details dictionary from decision records.

Creates a nested dictionary structure containing information about all duplicate relationships and decisions made during the duplicate review process.

Parameters:

Name Type Description Default
df_decisions DataFrame

DataFrame containing duplicate decisions with columns: - 'Decision 1', 'Decision 2': Decision for each record (KEEP/REMOVE/COMPOSITE) - 'datasetId 1', 'datasetId 2': IDs of the duplicate pair - 'originalDatabase 1', 'originalDatabase 2': Source databases - 'Decision type': Type of decision (MANUAL/AUTO) - 'Decision comment': Comments on the decision (for manual decisions)

 required
header list

Header information from the decision file containing operator details: [0]: File description [1]: Operator name and initials

[3]: Creation timestamp

 required

Returns:

Name Type Description
dup_details dict

Nested dictionary with structure: {record_id: { duplicate_count: { 'duplicate ID': str, 'duplicate database': str, 'duplicate decision': str, 'decision type': str, 'operator': str (for manual decisions), 'note': str (for manual decisions) } }}
Notes
  • Only processes pairs where at least one record is not kept (true duplicates)
  • Each record gets an entry for every duplicate relationship it has
  • Manual decisions include operator details extracted from the header
  • Automatic decisions have 'N/A' for operator and note fields

The returned dictionary can be used to populate the 'duplicateDetails' field in the final deduplicated database.

Source code in dod2k_utilities/ut_duplicate_search.py 
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
def collect_dup_details(df_decisions, header):
    """
    Generate duplicate details dictionary from decision records.

    Creates a nested dictionary structure containing information about all
    duplicate relationships and decisions made during the duplicate review process.

    Parameters
    ----------
    df_decisions : pandas.DataFrame
        DataFrame containing duplicate decisions with columns:
        - 'Decision 1', 'Decision 2': Decision for each record (KEEP/REMOVE/COMPOSITE)
        - 'datasetId 1', 'datasetId 2': IDs of the duplicate pair
        - 'originalDatabase 1', 'originalDatabase 2': Source databases
        - 'Decision type': Type of decision (MANUAL/AUTO)
        - 'Decision comment': Comments on the decision (for manual decisions)
    header : list
        Header information from the decision file containing operator details:
        [0]: File description
        [1]: Operator name and initials
        [2]: Email
        [3]: Creation timestamp

    Returns
    -------
    dup_details : dict
        Nested dictionary with structure:
        {record_id: {
            duplicate_count: {
                'duplicate ID': str,
                'duplicate database': str,
                'duplicate decision': str,
                'decision type': str,
                'operator': str (for manual decisions),
                'note': str (for manual decisions)
            }
        }}

    Notes
    -----
    - Only processes pairs where at least one record is not kept (true duplicates)
    - Each record gets an entry for every duplicate relationship it has
    - Manual decisions include operator details extracted from the header
    - Automatic decisions have 'N/A' for operator and note fields

    The returned dictionary can be used to populate the 'duplicateDetails' field
    in the final deduplicated database.
    """
    dup_details = {}
    dup_counts  = {}
    for ind in df_decisions.index:
        dec_1, dec_2 = df_decisions.loc[ind, ['Decision 1', 'Decision 2']]
        if dec_1=='KEEP' and dec_2=='KEEP': continue # in this case no true duplicates!

        id_1, id_2 = df_decisions.loc[ind, ['datasetId 1', 'datasetId 2']]
        db_1, db_2 = df_decisions.loc[ind, ['originalDatabase 1', 'originalDatabase 2']]
        for id in [id_1, id_2]:
            if id not in dup_details:
                dup_counts[id] = 0
                dup_details[id] = {}
            else:
                dup_counts[id]+=1
        dup_details[id_1][dup_counts[id_1]] = {'duplicate ID': id_2, 'duplicate database': db_2, 'duplicate decision': dec_2, 'decision type': df_decisions.loc[ind, 'Decision type']}
        dup_details[id_2][dup_counts[id_2]] = {'duplicate ID': id_1, 'duplicate database': db_1, 'duplicate decision': dec_1, 'decision type': df_decisions.loc[ind, 'Decision type']}

        if df_decisions.loc[ind, 'Decision type']=='MANUAL': 
            operator_details_str = ','.join(header[1:]).replace(' Modified ','')[:-2].replace(':','').replace('  E-Mail', '')
            for id in [id_1, id_2]:
                dup_details[id][dup_counts[id]]['operator'] = operator_details_str
                dup_details[id][dup_counts[id]]['note'] = df_decisions.loc[ind, 'Decision comment']
        else:
            for id in [id_1, id_2]:
                dup_details[id][dup_counts[id]]['operator'] = 'N/A'
                dup_details[id][dup_counts[id]]['note'] = 'N/A'       
    return dup_details

collect_record_decisions(df_decisions)

Collect per-record decisions from a pairwise decision table.

Parameters:

Name Type Description Default
df_decisions DataFrame

DataFrame containing pairwise record comparisons. Must include the columns 'datasetId 1', 'datasetId 2', 'Decision 1', and 'Decision 2', where each row represents a comparison between two dataset records and the associated decisions for each record.

 required

Returns:

Name Type Description
decisions dict

Dictionary mapping each dataset ID to a list of decisions associated with that record across all comparisons.
Notes

Each row contributes two entries: one for 'datasetId 1' paired with 'Decision 1' and one for 'datasetId 2' paired with 'Decision 2'. Records appearing in multiple rows accumulate multiple decision entries in their corresponding list.

Source code in dod2k_utilities/ut_duplicate_search.py 
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
def collect_record_decisions(df_decisions):
    """
    Collect per-record decisions from a pairwise decision table.

    Parameters
    ----------
    df_decisions : pandas.DataFrame
        DataFrame containing pairwise record comparisons. Must include
        the columns ``'datasetId 1'``, ``'datasetId 2'``,
        ``'Decision 1'``, and ``'Decision 2'``, where each row represents
        a comparison between two dataset records and the associated
        decisions for each record.

    Returns
    -------
    decisions : dict
        Dictionary mapping each dataset ID to a list of decisions
        associated with that record across all comparisons.

    Notes
    -----
    Each row contributes two entries: one for ``'datasetId 1'`` paired
    with ``'Decision 1'`` and one for ``'datasetId 2'`` paired with
    ``'Decision 2'``. Records appearing in multiple rows accumulate
    multiple decision entries in their corresponding list.
    """
    decisions = {}
    for ind in df_decisions.index:
        id1, id2   = df_decisions.loc[ind, ['datasetId 1', 'datasetId 2']]
        dec1, dec2 = df_decisions.loc[ind, ['Decision 1', 'Decision 2']]
        for id, dec in zip([id1, id2], [dec1, dec2]):
            if id not in decisions: decisions[id] = []
            decisions[id]+=[dec]
    return decisions

define_hierarchy(df, hierarchy='default')

Define priority hierarchy for different paleoclimate databases.

Assigns a numerical hierarchy value to each record based on its original database, used for automatic duplicate resolution when records are identical. Lower values indicate higher priority.

Parameters:

Name Type Description Default
df DataFrame

DataFrame containing proxy records with an 'originalDatabase' column.

 required
hierarchy str or dict

Hierarchy definition method. If 'default', uses predefined hierarchy. If dict, should contain keys for database names with priority values: - 'pages2k': priority for PAGES 2k database - 'fe23': priority for FE23 (Breitenmoser et al. 2014) - 'ch2k': priority for CoralHydro2k - 'iso2k': priority for Iso2k - 'sisal': priority for SISAL Default is 'default'.

 'default'

Returns:

Type Description
DataFrame

The input DataFrame with added 'Hierarchy' column containing priority values for each record.
Notes

Default hierarchy (lower number = higher priority): 1. PAGES 2k v2.2.0 (highest priority) 2. SISAL v3 3. CoralHydro2k v1.0.1 4. Iso2k v1.1.2 5. FE23 (Breitenmoser et al. 2014) 99. All other databases (lowest priority)

The hierarchy is used in duplicate_decisions functions to automatically choose which record to keep when duplicates are identical.

Examples:

>>> df = define_hierarchy(df)  # Use default hierarchy
>>> custom = {'PAGES 2k v2.2.0', 'Hierarchy': 2, 'SISAL v3': 1, 'FE23 (Breitenmoser et al. (2014))': 3, 'CoralHydro2k v1.0.': 4, 'Iso2k v1.1.2': 5}
>>> df = define_hierarchy(df, hierarchy=custom)  # Custom hierarchy
Source code in dod2k_utilities/ut_duplicate_search.py 
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
def define_hierarchy(df, hierarchy='default'):
    """
    Define priority hierarchy for different paleoclimate databases.

    Assigns a numerical hierarchy value to each record based on its original 
    database, used for automatic duplicate resolution when records are identical.
    Lower values indicate higher priority.

    Parameters
    ----------
    df : pandas.DataFrame
        DataFrame containing proxy records with an 'originalDatabase' column.
    hierarchy : str or dict, optional
        Hierarchy definition method. If 'default', uses predefined hierarchy.
        If dict, should contain keys for database names with priority values:
        - 'pages2k': priority for PAGES 2k database
        - 'fe23': priority for FE23 (Breitenmoser et al. 2014)
        - 'ch2k': priority for CoralHydro2k
        - 'iso2k': priority for Iso2k
        - 'sisal': priority for SISAL
        Default is 'default'.

    Returns
    -------
    pandas.DataFrame
        The input DataFrame with added 'Hierarchy' column containing priority
        values for each record.

    Notes
    -----
    Default hierarchy (lower number = higher priority):
    1. PAGES 2k v2.2.0 (highest priority)
    2. SISAL v3
    3. CoralHydro2k v1.0.1
    4. Iso2k v1.1.2
    5. FE23 (Breitenmoser et al. 2014)
    99. All other databases (lowest priority)

    The hierarchy is used in duplicate_decisions functions to automatically
    choose which record to keep when duplicates are identical.

    Examples
    --------
    >>> df = define_hierarchy(df)  # Use default hierarchy
    >>> custom = {'PAGES 2k v2.2.0', 'Hierarchy': 2, 'SISAL v3': 1, 'FE23 (Breitenmoser et al. (2014))': 3, 'CoralHydro2k v1.0.': 4, 'Iso2k v1.1.2': 5}
    >>> df = define_hierarchy(df, hierarchy=custom)  # Custom hierarchy
    """
    df['Hierarchy'] = 99
    if hierarchy=='default':
        df.loc[df['originalDatabase']=='PAGES 2k v2.2.0', 'Hierarchy'] = 1
        df.loc[df['originalDatabase']=='FE23 (Breitenmoser et al. (2014))', 'Hierarchy'] = 5
        df.loc[df['originalDatabase']=='CoralHydro2k v1.0.1', 'Hierarchy'] = 3
        df.loc[df['originalDatabase']=='Iso2k v1.1.2', 'Hierarchy'] = 4
        df.loc[df['originalDatabase']=='SISAL v3', 'Hierarchy'] = 2
    else:
        df.loc[df['originalDatabase']=='PAGES 2k v2.2.0', 'Hierarchy'] = hierarchy['pages2k']
        df.loc[df['originalDatabase']=='FE23 (Breitenmoser et al. (2014))', 'Hierarchy'] = hierarchy['fe23']
        df.loc[df['originalDatabase']=='CoralHydro2k v1.0.1', 'Hierarchy'] = hierarchy['ch2k']
        df.loc[df['originalDatabase']=='Iso2k v1.1.2', 'Hierarchy'] = hierarchy['iso2k']
        df.loc[df['originalDatabase']=='SISAL v3', 'Hierarchy'] = hierarchy['sisal']

    print('Chosen hierarchy:')



    for db, h in df.groupby('originalDatabase')['Hierarchy'].min().sort_values().items():
        if h==1: 
            print(f'{h}. {db} (highest)')
        elif h==len(df.originalDatabase.unique()): 
            print(f'{h}. {db} (lowest)')
        else:
            print(f'{h}. {db}')
    return df

density_scatter(x, y, ax=None, fig=None, sort=True, bins=20, **kwargs)

Scatter plot colored by 2d histogram

Source code in dod2k_utilities/ut_duplicate_search.py 
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
def density_scatter( x , y, ax = None, fig=None, sort = True, bins = 20, **kwargs )   :
    """
    Scatter plot colored by 2d histogram
    """

    from matplotlib import cm
    from matplotlib.colors import Normalize 
    from scipy.interpolate import interpn
    if ax is None :
        fig , ax = plt.subplots()
    data , x_e, y_e = np.histogram2d( x, y, bins = bins, density = True )
    z = interpn( ( 0.5*(x_e[1:] + x_e[:-1]) , 0.5*(y_e[1:]+y_e[:-1]) ) , data , np.vstack([x,y]).T , method = "splinef2d", bounds_error = False)

    #To be sure to plot all data
    z[np.where(np.isnan(z))] = 0.0

    # Sort the points by density, so that the densest points are plotted last
    if sort :
        idx = z.argsort()
        x, y, z = x[idx], y[idx], z[idx]

    ax.scatter( x, y, c=z, **kwargs )

    norm = Normalize(vmin = np.min(z), vmax = np.max(z))
    cbar = plt.colorbar(cm.ScalarMappable(norm = norm), ax=ax)
    cbar.ax.set_ylabel('data density')

    return ax

dup_plot(df, ii, jj, id_1, id_2, time_1, time_2, time_12, data_1, data_2, int_1, int_2, pot_dup_corr, keys_to_print=['originalDatabase', 'originalDataURL', 'datasetId', 'archiveType', 'proxy | variableName', 'geo_siteName', 'lat | lon | elev', 'mean | std | units', 'year'], dup_mdata_row=[], plot_text=True, fig_scale=1)

Plots the duplicate candidates. Plots the record data as a timeseries of anomalies in a common panel (w.r.t. shared time period) and prints out the most relevant metadata. Highlights identical metadata in orange and different metadata in green.

Parameters:

Name Type Description Default
df DataFrame

DataFrame containing proxy records and metadata.

 required
ii int

Index of the first record.

 required
jj int

Index of the second record.

 required
id_1 str

Dataset ID of the first record.

 required
id_2 str

Dataset ID of the second record.

 required
time_1 array - like

Time vector for the first record.

 required
time_2 array - like

Time vector for the second record.

 required
time_12 array - like

Shared time points between both records.

 required
data_1 array - like

Data values of the first record.

 required
data_2 array - like

Data values of the second record.

 required
int_1 array - like

Indices of shared times in the first record.

 required
int_2 array - like

Indices of shared times in the second record.

 required
pot_dup_corr float

Correlation between the two records.

 required
keys_to_print list of str

List of metadata keys to display. Default is a standard set of keys.

 ['originalDatabase', 'originalDataURL', 'datasetId', 'archiveType', 'proxy | variableName', 'geo_siteName', 'lat | lon | elev', 'mean | std | units', 'year']
dup_mdata_row list

Stores metadata differences for output.

 []

Returns:

Name Type Description
fig Figure

Figure object of the duplicate plot.
dup_mdata_row list

Metadata rows generated for display.
Source code in dod2k_utilities/ut_duplicate_search.py 
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
def dup_plot(df, ii, jj, id_1, id_2, time_1, time_2, time_12, data_1, data_2, int_1, int_2, 
             pot_dup_corr, keys_to_print=['originalDatabase', 'originalDataURL', 'datasetId', 'archiveType', 
                                          'proxy | variableName', #'archive|proxy', 
                                          'geo_siteName', 'lat | lon | elev', 'mean | std | units', 'year' ], 
             dup_mdata_row=[], plot_text=True, fig_scale=1):
    """
    Plots the duplicate candidates. Plots the record data as a timeseries of anomalies in a common panel (w.r.t. shared time period) and prints out the most relevant metadata. Highlights identical metadata in orange and different metadata in green.

    Parameters
    ----------
    df : pandas.DataFrame
        DataFrame containing proxy records and metadata.
    ii : int
        Index of the first record.
    jj : int
        Index of the second record.
    id_1 : str
        Dataset ID of the first record.
    id_2 : str
        Dataset ID of the second record.
    time_1 : array-like
        Time vector for the first record.
    time_2 : array-like
        Time vector for the second record.
    time_12 : array-like
        Shared time points between both records.
    data_1 : array-like
        Data values of the first record.
    data_2 : array-like
        Data values of the second record.
    int_1 : array-like
        Indices of shared times in the first record.
    int_2 : array-like
        Indices of shared times in the second record.
    pot_dup_corr : float
        Correlation between the two records.
    keys_to_print : list of str, optional
        List of metadata keys to display. Default is a standard set of keys.
    dup_mdata_row : list, optional
        Stores metadata differences for output.

    Returns
    -------
    fig : matplotlib.figure.Figure
        Figure object of the duplicate plot.
    dup_mdata_row : list
        Metadata rows generated for display.
    """
    from matplotlib import font_manager
    fs        = 10
    scale     = 0.16
    init_offs = -.6
    str_limit = 50

    fig  = plt.figure(figsize=(10, 8.5), dpi=300*fig_scale)
    grid = GS(3,4, wspace=0.6)
    ax   = plt.subplot(grid[0,:3])
    #
    label1 = f'#{ii}: {id_1}' 
    # print('int_1', int_1)
    ax.scatter(time_1, data_1-np.mean(data_1[int_1]), facecolor='None', 
               edgecolor='tab:blue', marker='o',#marker='o',
               s=15, label=label1, alpha=0.9)

    label2 = f'#{jj}: {id_2}'# f'ID2: {id_2} (#{jj}))' 
    ax.scatter(time_2, data_2-np.mean(data_2[int_2]), color='tab:red', marker='x',#marker='D',
               s=15, label=label2, alpha=0.9)


    ax.legend(loc='upper right')
    ax.set_ylabel(f'anomalies w.r.t. {int(time_12[0])}-{int(time_12[-1])} \n '+
                  f'#{ii}: '+str(df['paleoData_units'].iloc[ii])+f'#{jj}: '+str(df['paleoData_units'].iloc[jj]))

    ax.set_xlabel(df['yearUnits'].iloc[ii])
    ax.set_title('Possible duplicates. Correlation=%3.2f'%pot_dup_corr
                 +'. Time overlap=%d (%d'%(len(time_12), len(time_12)/len(time_1)*100)+
                 '%'+', %d'%(len(time_12)/len(time_2)*100)+'%)')

    txt = ('Metadata (differences are highlighted in blue; only first '+
           str(str_limit)+' characters are displayed)')

    # column positions
    cols = [.1, -.1+0, -.1+.3, -.1+.9+0.15]

    rows = [-0.3]+[-0.45]*3

    ax.text(cols[0], rows[0], txt, transform=ax.transAxes, ha='left', fontsize=fs-1)

    ax.text(cols[1], rows[1],'Key', transform=ax.transAxes, fontsize=fs+2, ha='left')
    ax.text(cols[2], rows[2],'Record 1 (blue circles)', transform=ax.transAxes, ha='left', fontsize=fs+1)
    ax.text(cols[3], rows[3],'Record 2 (red crosses)',  transform=ax.transAxes, ha='left', fontsize=fs+1)
    # plot the metadata 
    for ik, key in enumerate(keys_to_print):
        if key=='lat | lon | elev':
            metadata_field1 = '%3.1f | %3.1f | %3.1f'%(df['geo_meanLat'].iloc[ii],
                                                   df['geo_meanLon'].iloc[ii], 
                                                   df['geo_meanElev'].iloc[ii])
            metadata_field2 = '%3.1f | %3.1f | %3.1f'%(df['geo_meanLat'].iloc[jj],
                                                   df['geo_meanLon'].iloc[jj], 
                                                   df['geo_meanElev'].iloc[jj])
        elif key=='year':
            metadata_field1 = '%3.1f - %3.1f %s'%(np.min(df['year'].iloc[ii]), 
                                                  np.max(df['year'].iloc[ii]), 
                                                  df['yearUnits'].iloc[ii])
            metadata_field2 = '%3.1f - %3.1f %s'%(np.min(df['year'].iloc[jj]), 
                                                  np.max(df['year'].iloc[jj]), 
                                                  df['yearUnits'].iloc[jj])
        elif key=='mean | std | units':
            metadata_field1 = '%4.2f | %4.2f | %s'%(np.mean(df['paleoData_values'].iloc[ii]), 
                                                  np.std(df['paleoData_values'].iloc[ii]), 
                                                  df['paleoData_units'].iloc[ii])
            metadata_field2 = '%4.2f | %4.2f | %s'%(np.mean(df['paleoData_values'].iloc[jj]), 
                                                  np.std(df['paleoData_values'].iloc[jj]), 
                                                  df['paleoData_units'].iloc[jj])
        elif key=='archive | proxy': #'archiveType','paleoData_proxy',
            metadata_field1 = '%s | %s'%(df['archiveType'].iloc[ii],
                                       df['paleoData_proxy'].iloc[ii])
            metadata_field2 = '%s | %s'%(df['archiveType'].iloc[jj],
                                       df['paleoData_proxy'].iloc[jj])
        elif key=='proxy | variableName': #'archiveType','paleoData_proxy',
            metadata_field1 = '%s | %s'%(df['paleoData_proxy'].iloc[ii],
                                       df['paleoData_variableName'].iloc[ii])
            metadata_field2 = '%s | %s'%(df['paleoData_proxy'].iloc[jj],
                                       df['paleoData_variableName'].iloc[jj])
        else:
            metadata_field1, metadata_field2 = df[key].iloc[[ii, jj]]

        if 'URL' in key:
            metadata_field1 = metadata_field1.replace('https://','')
            metadata_field2 = metadata_field2.replace('https://','')

        row = init_offs-scale*ik
        col = [-.1, -.1+.2, -.1+.75]



        plt.text(cols[1], row, ' '*180,transform=ax.transAxes, fontsize=fs+1, ha='left', 
                 bbox={'facecolor': 'tab:grey',   'alpha':.05, 'pad':2.5, 'edgecolor': 'None'
                      })
        # if metadata_field1 != metadata_field2:
        #     c = 'white'
        #     # plt.text(cols[1], row, ' '*250,transform=ax.transAxes, fontsize=fs+1, ha='left', 
        #     #          bbox={'facecolor': 'tab:blue', 
        #     #                'alpha':.2, 'pad':2.5, 'edgecolor': 'None'
        #     #               })
        # else:
        #     c = 'tab:orange'
        #     # plt.text(cols[1], row, ' '*250,transform=ax.transAxes, fontsize=fs+1, ha='left',
        #     #          bbox={'facecolor':'tab:orange','alpha':.2, 'pad':2.5, 'edgecolor': 'None'})

        # ax.text(cols[1], row, key, transform=ax.transAxes,fontsize=fs, ha='left', 
        #         bbox={'facecolor':c,'alpha':.2, 'pad':2.5, 'edgecolor': 'None'})
        # ax.text(cols[2], row, str(metadata_field1)[0:str_limit],ha='left', 
        #         transform=ax.transAxes,fontsize=fs, 
        #         bbox={'facecolor':c,'alpha':.2, 'pad':2.5, 'edgecolor': 'None'})
        # ax.text(cols[3], row, str(metadata_field2)[0:str_limit],ha='left', 
        #         transform=ax.transAxes,fontsize=fs, 
        #         bbox={'facecolor':c,'alpha':.2, 'pad':2.5, 'edgecolor': 'None'})


        # In the metadata loop, replace the current text plotting with:



        # if metadata_field1 != metadata_field2:
            # Plot background for the whole row
        plt.text(cols[1], row, ' '*180, transform=ax.transAxes, fontsize=fs+1, ha='left', 
                 bbox={'facecolor': 'tab:grey', 'alpha':.05, 'pad':2.5, 'edgecolor': 'None'})

        # For compound fields, split and highlight individually

        if '|' in str(metadata_field1):# Plot background for the whole row
            plt.text(cols[1], row, ' '*180, transform=ax.transAxes, fontsize=fs+1, ha='left', 
                     bbox={'facecolor': 'tab:grey', 'alpha':.05, 'pad':2.5, 'edgecolor': 'None'})


            parts1 = str(metadata_field1).split('|')
            parts2 = str(metadata_field2).split('|')

            # Plot the key normally
            ax.text(cols[1], row, key, transform=ax.transAxes, fontsize=fs, #family='monospace',
                    ha='left')

            # Calculate spacing for each part
            cumulative_offset = 0
            char_width = 0.015  # Approximate character width in axes coordinates (adjust as needed)

            for i, (p1, p2) in enumerate(zip(parts1, parts2)):
                p1, p2 = p1.strip(), p2.strip()
                color = 'tab:orange' if p1 == p2 else 'white'

                # For record 1 (cols[2])
                ax.text(cols[2] + cumulative_offset, row, p1, ha='left',#family='monospace',
                       transform=ax.transAxes, fontsize=fs,
                       bbox={'facecolor': color, 'alpha': .2, 'pad': 2.5, 'edgecolor': 'None'})

                # For record 2 (cols[3])
                ax.text(cols[3] + cumulative_offset, row, p2, ha='left',#family='monospace',
                       transform=ax.transAxes, fontsize=fs,
                       bbox={'facecolor': color, 'alpha': .2, 'pad': 2.5, 'edgecolor': 'None'})

                # Add separator if not last element
                if i < len(parts1) - 1:
                    sep_offset = cumulative_offset + len(p1) * char_width
                    ax.text(cols[2] + sep_offset, row, ' | ', ha='left',#family='monospace',
                           transform=ax.transAxes, fontsize=fs)
                    ax.text(cols[3] + sep_offset, row, ' | ', ha='left',#family='monospace',
                           transform=ax.transAxes, fontsize=fs)
                    cumulative_offset = sep_offset + 3 * char_width
                else:
                    cumulative_offset += len(p1) * char_width

        else:
            if metadata_field1 != metadata_field2:
                # Non-compound field - use original logic
                c = 'white'
                ax.text(cols[1], row, key, transform=ax.transAxes, fontsize=fs, ha='left', #family='monospace',
                        bbox={'facecolor': c, 'alpha': .2, 'pad': 2.5, 'edgecolor': 'None'})
                ax.text(cols[2], row, str(metadata_field1)[0:str_limit], ha='left', #family='monospace',
                        transform=ax.transAxes, fontsize=fs,
                        bbox={'facecolor': c, 'alpha': .2, 'pad': 2.5, 'edgecolor': 'None'})
                ax.text(cols[3], row, str(metadata_field2)[0:str_limit], ha='left',#family='monospace',
                        transform=ax.transAxes, fontsize=fs,
                        bbox={'facecolor': c, 'alpha': .2, 'pad': 2.5, 'edgecolor': 'None'})
            else:
                # Fields match - keep original orange highlighting
                c = 'tab:orange'
                plt.text(cols[1], row, ' '*180, transform=ax.transAxes, fontsize=fs+1, ha='left', #family='monospace',
                         bbox={'facecolor': 'tab:grey', 'alpha':.05, 'pad':2.5, 'edgecolor': 'None'})
                ax.text(cols[1], row, key, transform=ax.transAxes, fontsize=fs, ha='left',#family='monospace',
                        bbox={'facecolor': 'None', 'alpha': .2, 'pad': 2.5, 'edgecolor': 'None'})
                ax.text(cols[2], row, str(metadata_field1)[0:str_limit], ha='left',#family='monospace',
                        transform=ax.transAxes, fontsize=fs,
                        bbox={'facecolor': c, 'alpha': .2, 'pad': 2.5, 'edgecolor': 'None'})
                ax.text(cols[3], row, str(metadata_field2)[0:str_limit], ha='left',#family='monospace',
                        transform=ax.transAxes, fontsize=fs,
                        bbox={'facecolor': c, 'alpha': .2, 'pad': 2.5, 'edgecolor': 'None'})

        dup_mdata_row += [[metadata_field1, metadata_field2][m_ind].split('|')[s_ind]   
                          for s_ind in range(len(metadata_field1.split('|'))) 
                          for m_ind in [0,1]]
    ax2   = plt.subplot(grid[0,3])

    x  = data_1[int_1]-np.mean(data_1[int_1])
    y  = data_2[int_2]-np.mean(data_2[int_2])

    # density_scatter(x , y, ax = ax2, fig=fig)

    # xy = np.vstack([x,y])
    # z  = gaussian_kde(xy)(xy)

    # idx = z.argsort()
    # x, y, z = x[idx], y[idx], z[idx]


    density_scatter(x , y, ax2, s=5)
    # ax2.scatter(x, y, color='k', marker='+', alpha=0.5)#,
    #             #s=15, linestyle='None')
    plt.xlabel(label1)
    plt.ylabel(label2)

    # Add 1:1 line spanning the data range
    min_val = min(x.min(), y.min())
    max_val = max(x.max(), y.max())
    ax2.plot([min_val, max_val], [min_val, max_val],color='tab:grey',  
             alpha=0.75, zorder=0, label='1:1 line')


    return fig, dup_mdata_row

duplicate_decisions_multiple(df, operator_details=False, choose_recollection=True, plot=True, remove_identicals=True, dist_tolerance_km=8, backup=True, comment=True, automate_db_choice=False)

Review potential duplicate pairs in a proxy database and decide which records to keep, remove, or combine.

This function walks through each potential duplicate pair identified in a dataset, displays metadata and optionally plots the data, and allows the operator to make decisions. Decisions are saved to a CSV file, and a duplicate-free dataframe can be generated.

Copied from duplicate_decisions but improved handling of multiple duplicates.

Parameters:

Name Type Description Default
df DataFrame

The dataframe containing proxy data and metadata. Expected columns include: - 'geo_meanLat', 'geo_meanLon' : Latitude and longitude of the site. - 'year', 'paleoData_values' : Time vector and proxy values. - 'archiveType', 'paleoData_proxy' : Archive type and proxy type. - 'datasetId' : Unique identifier or dataset name. - 'geo_siteName' : Site name. - 'originalDatabase' : Name of the original database. - 'geo_meanElev' : Mean elevation of the site (optional for duplicate checks). - 'Hierarchy' : Numeric value representing dataset priority (used for auto-decisions). - 'originalDataURL' : URL of the original dataset.

 required
operator_details tuple or bool

Tuple containing (initials, fullname, email) of the operator. If False, user input is requested. Default is False.

 False
choose_recollection bool

If True, automatically selects the record that is a recollection or update when applicable. Default is True.

 True
plot bool

If True, generate plots for manual inspection of duplicate pairs. Default is True.

 True
remove_identicals bool

If True, automatically remove records that are identical in data and metadata. Default is True.

 True
dist_tolerance_km float

Maximum distance (in km) for considering records as recollection updates. Default is 8 km.

 8

Returns:

Type Description
None

Decisions are saved as CSV backup and final CSV in the df.name/dup_detection/ directory.
Notes
  • Automatic decisions are made based on data identity, metadata identity, perfect correlation, and recollection indicators in site names.
  • Manual decisions are prompted via command-line input if automatic rules do not apply.
  • Decision types include:
    • 'AUTO: UPDATE' : Automatically select record that is a recollection/update.
    • 'AUTO: IDENTICAL' : Automatically select record based on hierarchy if records are identical.
    • 'MANUAL' : Decision requires operator input.
  • Figures are saved with a standardized naming convention and linked in the CSV output.
Source code in dod2k_utilities/ut_duplicate_search.py 
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
def duplicate_decisions_multiple(df, operator_details=False, choose_recollection=True, #keep_all=False, 
                                 plot=True, remove_identicals=True, dist_tolerance_km=8, backup=True,
                                 comment=True, automate_db_choice=False):
    """
    Review potential duplicate pairs in a proxy database and decide which records to keep, remove, or combine.

    This function walks through each potential duplicate pair identified in a dataset,
    displays metadata and optionally plots the data, and allows the operator to make decisions.
    Decisions are saved to a CSV file, and a duplicate-free dataframe can be generated.

    Copied from duplicate_decisions but improved handling of multiple duplicates.

    Parameters
    ----------
    df : pd.DataFrame
        The dataframe containing proxy data and metadata. Expected columns include:
            - 'geo_meanLat', 'geo_meanLon' : Latitude and longitude of the site.
            - 'year', 'paleoData_values' : Time vector and proxy values.
            - 'archiveType', 'paleoData_proxy' : Archive type and proxy type.
            - 'datasetId' : Unique identifier or dataset name.
            - 'geo_siteName' : Site name.
            - 'originalDatabase' : Name of the original database.
            - 'geo_meanElev' : Mean elevation of the site (optional for duplicate checks).
            - 'Hierarchy' : Numeric value representing dataset priority (used for auto-decisions).
            - 'originalDataURL' : URL of the original dataset.

    operator_details : tuple or bool, optional
        Tuple containing (initials, fullname, email) of the operator. If False, user input is requested.
        Default is False.
    choose_recollection : bool, optional
        If True, automatically selects the record that is a recollection or update when applicable.
        Default is True.
    plot : bool, optional
        If True, generate plots for manual inspection of duplicate pairs. Default is True.
    remove_identicals : bool, optional
        If True, automatically remove records that are identical in data and metadata. Default is True.
    dist_tolerance_km : float, optional
        Maximum distance (in km) for considering records as recollection updates. Default is 8 km.

    Returns
    -------
    None
        Decisions are saved as CSV backup and final CSV in the `df.name/dup_detection/` directory.

    Notes
    -----
    - Automatic decisions are made based on data identity, metadata identity, perfect correlation,
      and recollection indicators in site names.
    - Manual decisions are prompted via command-line input if automatic rules do not apply.
    - Decision types include:
        - 'AUTO: UPDATE' : Automatically select record that is a recollection/update.
        - 'AUTO: IDENTICAL' : Automatically select record based on hierarchy if records are identical.
        - 'MANUAL' : Decision requires operator input.
    - Figures are saved with a standardized naming convention and linked in the CSV output.
    """

    def create_ID_dup_dict(pot_dup_IDs):
        dup_dict         = {}
        reverse_dup_dict = {}
        for id1, id2 in pot_dup_IDs:
            # check if id1 appears in reverse_dup_dict: 
            if id1 in reverse_dup_dict:
                dup_id = reverse_dup_dict[id1] # if YES, it means it is already associated with a duplicate record, use this for mapping
            else:
                dup_id = id1 # if NO this is the first time id1 appears as a potential dup, then use this for mapping
            # dup_id is the unique mapper to the duplicate category.
            if dup_id not in dup_dict: 
                dup_dict[dup_id] = [] # create an empty dictionary key in case this is the first time that dup_id appears
            # if dup_id!=id1: # if id1 is not identical to the unique mapper
            #     if id1 not in dup_dict[dup_id]:
            #         dup_dict[dup_id]+=[id1]
            if id1 not in dup_dict[dup_id]:
                if dup_id!=id1:
                    dup_dict[dup_id]+=[id1] # add id1 to the dup_dict if not in there yet (keep entries unique) and id1 is not unique mapper
            if id2 not in dup_dict[dup_id]:
                dup_dict[dup_id]+=[id2]  # add id2 to the dup_dict if not in there yet (keep entries unique)

            reverse_dup_dict[id1] = dup_id # make sure id1 gets mapped back to its unique mapper
            reverse_dup_dict[id2] = dup_id # make sure id2 gets mapped back to its unique mapper


        # # drop all PAIRS to get multiples only
        dup_dict_multiples = {}
        for kk, vv in dup_dict.items():
            if len(vv)>1:
                dup_dict_multiples[kk]=vv
        multiple_list = []
        for kk, vv in dup_dict_multiples.items():
            # print(kk, len(vv), vv)
            multiple_list.append(kk)
            for vvv in vv:
                multiple_list.append(vvv)
        print('------'*10)
        if len(dup_dict_multiples)>0:
            print('Detected MULTIPLE duplicates, including:')
            for kk, vv in dup_dict_multiples.items():
                print(kk, len(vv), vv)
            print('PLEASE PAY ATTENTION WHEN MAKING DECISIONS FOR THESE DUPLICATES!')
            print('The decision process will go through the duplicates on a PAIR-BY-PAIR basis, which is not optimised for multiple duplicates.')
            print('The multiples will be highlighted throughout the decision process.')
            print('Should the operator want to go back and revise a previous decision based on the presentation of a new candidate pair, they can manually modify the backup file to alter any previous decisions.')
        print('------'*10)
        return dup_dict_multiples, reverse_dup_dict, multiple_list 

    import datetime
    # Select the metadata keys to show on the output figures - keys depend on the 
    # dataframe and need to be determined according to the dataframe input. 
    # if using this code on fused_database.pkl, make sure that all entries are available 
    # in the dataframe1 Use fuse_datasets.py to fuse and homogenise metadata  of different databases.

    if not operator_details:
        initials = input('Please enter your initials here:')
        fullname = input('Please enter your full name here:')
        email    = input('Please enter your email address here:')
    else:
        initials, fullname, email = operator_details
    date_time= str(datetime.datetime.utcnow())+' (UTC)'
    header   = [['# Decisions for duplicate candidate pairs. ']]
    header  += [['# Operated by %s (%s)'%(fullname, initials)]]
    header  += [['# E-Mail: %s'%email]]
    header  += [['# Created on: %s'%date_time]]


    dirname = 'data/%s/dup_detection/'%(df.name)
    filename = f'dup_decisions_{df.name}' # name of csv file which saves duplicate candidate pairs

    filename+=f'_{initials}'

    detection_file = dirname+'dup_detection_candidates_'+df.name


    if automate_db_choice is not False:
        auto_db_choice = True
        auto_db_pref = automate_db_choice['preferred_db']
        auto_db_rej  = automate_db_choice['rejected_db']
    else:
        auto_db_choice = False
        auto_db_pref = ''
        auto_db_rej = ''


    if not os.path.exists(dirname):
        os.makedirs(dirname)

    # try to load data from backup!
    try:
        data, hh = read_csv(dirname+filename+'_BACKUP', header=True, last_header_row=4)
        if backup:
            backup_file=input(f'Found backup file ({dirname+filename}_BACKUP.csv). Do you want to start decision process from the backup file? [y/n]')
        else:
            backup_file='n'
        if backup_file=='n':
            raise FileNotFoundError
        data, hh = read_csv(dirname+filename+'_BACKUP', header=True, last_header_row=4)
        print('header', hh)
        print('data', list(data))
        data = list(data)

        load_saved_data      = True
        last_index = len(data)
        print('start with index: ', last_index)
    except FileNotFoundError:
        print('No back up.')
        load_saved_data      = False
        data = []
        pass

    # load the potential duplicate data as found in find_duplicates function
    pot_dup_meta, head = read_csv(detection_file, header=True)

    pot_dup_inds   = np.array(np.array(pot_dup_meta)[:, :2], dtype=int)  # indices for each pairs
    pot_dup_IDs    = np.array(np.array(pot_dup_meta)[:, 2:4], dtype=str) # IDs for each pairs
    pot_dup_corrs  = np.array(np.array(pot_dup_meta)[:, 4], dtype=float)
    pot_dup_dists  = np.array(np.array(pot_dup_meta)[:, 5], dtype=float)


    dup_dict_multiples, reverse_dict, multiple_list   = create_ID_dup_dict(pot_dup_IDs)
    # raise Exception

    n_pot_dups   = pot_dup_inds.shape[0]

    # loop through the potential duplicates.
    decisions = ['N/A']*df.shape[0] #set False if index should be discarded from dataset.
    cols    = [['index 1', 'index 2', 'figure path',
                'datasetId 1', 'datasetId 2', 
                'originalDatabase 1', 'originalDatabase 2',
                'geo_siteName 1', 'geo_siteName 2', 
                'geo_meanLat 1', 'geo_meanLat 2', 
                'geo_meanLon 1', 'geo_meanLon 2', 
                'geo_meanElevation 1', 'geo_meanElevation 2', 
                'archiveType 1', 'archiveType 2',
                'paleoData_proxy 1', 'paleoData_proxy 2',
                'originalDataURL 1', 'originalDataURL 2',
                'year 1', 'year 2',
                'Decision 1', 'Decision 2',
                'Decision type', 'Decision comment' ]]
    dup_dec = []


    # Write header and existing data ONCE before the loop (only if starting fresh)
    if not load_saved_data:
        # start a new backup file. 
        with open(dirname+filename+'_BACKUP.csv', 'w', newline='') as f:
            writer = csv.writer(f)
            writer.writerows(header)
            writer.writerows(cols)

    else:
        # If loading from backup, populate dup_dec with existing data
        for data_row in list(data):
            dup_dec += [list(data_row)]

    # Open backup file in APPEND mode for writing new decisions
    with open(dirname+filename+'_BACKUP.csv', 'a', newline='') as f:
        writer = csv.writer(f)

        for i_pot_dups, (ii, jj) in enumerate(pot_dup_inds):
            if load_saved_data:
                if i_pot_dups<last_index: continue
                else:
                    print(ii, jj, last_index, i_pot_dups)
            # data and metadata associated with the two potential duplicates
            id_1, id_2       = df['datasetId'].loc[[ii, jj]]
            time_1, time_2   = df['year'].loc[[ii, jj]]
            time_12, int_1, int_2 = np.intersect1d(time_1, time_2, return_indices=True) 
            data_1           = np.array(df['paleoData_values'].loc[ii])
            data_2           = np.array(df['paleoData_values'].loc[jj])

            lat_1, lat_2     = df['geo_meanLat'].loc[[ii, jj]]
            lon_1, lon_2     = df['geo_meanLon'].loc[[ii, jj]]
            site_1, site_2   = df['geo_siteName'].loc[[ii, jj]]
            or_db_1, or_db_2 = df['originalDatabase'].loc[[ii, jj]]

            auto_choice = '1' if df['Hierarchy'].loc[ii]>=df['Hierarchy'].loc[jj] else '2'

            print('> %d/%d'%(i_pot_dups+1, n_pot_dups), #set_txt, 
                   id_1, id_2, #var_1, var_2, season_1,season_2,
                   pot_dup_dists[i_pot_dups], pot_dup_corrs[i_pot_dups],
                   sep=',')

            # Print the interpretation values 
            print('====================================================================')
            print('=== POTENTIAL DUPLICATE %d/%d'%(i_pot_dups, n_pot_dups)+': %s+%s ==='%(df['datasetId'].loc[ii], df['datasetId'].loc[jj]))
            print('=== URL 1: %s   ==='%(df['originalDataURL'].loc[ii]))
            print('=== URL 2: %s   ==='%(df['originalDataURL'].loc[jj]))

            keep = ''
            dec_comment = ''


            elevation_nan = (np.isnan(df['geo_meanElev'].loc[ii])|
                             np.isnan(df['geo_meanElev'].loc[jj]))

            metadata_identical = ((np.abs(lat_1-lat_2)<=0.1) & 
                                  (np.abs(lon_1-lon_2)<=0.1)  & 
                                  ((np.abs(df['geo_meanElev'].loc[ii]-df['geo_meanElev'].loc[jj])<=1) 
                                    | elevation_nan) & 
                                  (df['archiveType'].loc[ii]==df['archiveType'].loc[jj]) & 
                                  (df['paleoData_proxy'].loc[ii]==df['paleoData_proxy'].loc[jj]) 
                                 )
            sites_identical     = site_1==site_2
            URL_identical       = (df['originalDataURL'].loc[ii]==df['originalDataURL'].loc[jj]) 
            # print(data_1, data_2)
            data_identical      = (list(data_1)==list(data_2)) & (list(time_1)==list(time_2))
            correlation_perfect = (True if pot_dup_corrs[i_pot_dups]>=0.98 else False) & (len(time_1)==len(time_2))

            # if one db is the preferred and the other the rejected db, only if automate_db_choice is not False
            auto_db_choice_cond = auto_db_choice & (((or_db_1==auto_db_pref) & (or_db_2==auto_db_rej))|((or_db_2==auto_db_pref) & (or_db_1==auto_db_rej)))

            print('True if pot_dup_corrs[i_pot_dups]>=0.98 else False', True if pot_dup_corrs[i_pot_dups]>=0.98 else False)
            print('(len(time_1)==len(time_2))', (len(time_1)==len(time_2)))

            print('metadata_identical: ', metadata_identical)
            print('lat', (np.abs(lat_1-lat_2)<=0.1), 'lon' , (np.abs(lon_1-lon_2)<=0.1)   ,'elevation',
                                  ( (np.abs(df['geo_meanElev'].loc[ii]-df['geo_meanElev'].loc[jj])<=1) 
                                    | elevation_nan) , 'archivetype',
                                  (df['archiveType'].loc[ii]==df['archiveType'].loc[jj]) ,'paleodata_proxy',
                                  (df['paleoData_proxy'].loc[ii]==df['paleoData_proxy'].loc[jj]) 
                                 )
            print('sites_identical: ', sites_identical)
            print('URL_identical: ', URL_identical)
            print('data_identical: ', data_identical)
            print('correlation_perfect: ', correlation_perfect)
            figpath = 'no figure'
            while keep not in ['1', '2', 'b', 'n', 'c']:
                # go through the data and metadata and check if decision is =automatic or manual
                if (choose_recollection 
                    # if record 1 is update of record 2, choose 1
                      & np.any([(ss in site_1.lower()) for ss in ['recollect', 'update', 're-collect']]) 
                      & (pot_dup_dists[i_pot_dups]<dist_tolerance_km)
                      & ~np.any([ss in site_2.lower() for ss in ['recollect', 'update', 're-collect']])
                     ):
                    keep = '1'
                    dec_comment = 'Record 1 (%s) is UPDATE of record 2(%s) . Automatically choose 1.'%(id_1, id_2)
                    print(dec_comment)
                    dec_type='AUTO: UPDATE'
                elif (choose_recollection 
                      # if record 2 is update of record 1, choose 2
                      & np.any([ss in site_2.lower() for ss in ['recollect', 'update', 're-collect']]) 
                      & (pot_dup_dists[i_pot_dups]<dist_tolerance_km)
                      & ~np.any([ss in site_1.lower() for ss in ['recollect', 'update', 're-collect']])
                     ):
                    keep = '2'
                    dec_comment = 'Record 2 (%s) is UPDATE of record 1 (%s). Automatically choose 2.'%(id_2, id_1)
                    print(dec_comment)
                    dec_type='AUTO: UPDATE'
                elif (remove_identicals & metadata_identical & sites_identical & URL_identical & (data_identical|correlation_perfect)):
                    # if all metadata and data matches, choose record according to hierarchy of databases
                    if data_identical:
                        dec_comment = 'RECORDS IDENTICAL (identical data). Automatically choose #%s.'%auto_choice
                    else:
                        dec_comment = 'RECORDS IDENTICAL (perfect correlation). Automatically choose #%s.'%auto_choice

                    print(dec_comment)
                    keep = auto_choice#'1'
                    dec_type='AUTO: IDENTICAL'
                elif (remove_identicals & (data_identical|correlation_perfect)):
                    # if most metadata and data matches except URL and/or site name, choose record according to hierarchy of databases
                    if data_identical:
                        dec_comment = 'RECORDS IDENTICAL (identical data) except for metadata. Automatically choose #%s.'%auto_choice
                    else:
                        dec_comment = 'RECORDS IDENTICAL (perfect correlation) except for metadata. Automatically choose #%s.'%auto_choice

                    print(dec_comment)
                    keep = auto_choice 
                    dec_type='AUTO: IDENTICAL except for URLs and/or geo_siteName.'
                elif auto_db_choice_cond & metadata_identical:
                    reason = automate_db_choice['reason']
                    dec_comment = f'Automated choice. Metadata identical, automatically choose {auto_db_pref} over {auto_db_rej}. {reason}'

                    print(dec_comment)
                    if or_db_1==auto_db_pref:
                        keep     =  '1' 
                    elif or_db_2==auto_db_pref:
                        keep     =  '2' 
                    else:
                        raise Exception('Auto condition error.')
                    dec_type = 'AUTO: preferred db and metadata identical.'

                else:

                    dec_type = 'MANUAL'
                    fig, dup_mdata_row = dup_plot(df, ii, jj, id_1, id_2, time_1, time_2, time_12, data_1, data_2, int_1, int_2, pot_dup_corrs[i_pot_dups])
                    plt.show(block=False)
                    print('**Decision required for this duplicate pair (see figure above).**')

                    if (id_1 in multiple_list) | (id_2 in multiple_list):
                        print('-------'*15)
                        print('***ATTENTION*** THIS RECORD IS ASSOCIATED WITH MULTIPLE DUPLICATES! PLEASE PAY SPECIAL ATTENTION WHEN MAKING DECISIONS FOR THIS RECORD!')
                        print('The potential duplicates also associated with this record are:')
                        if id_1!=reverse_dict[id_1]:
                            print('- ', reverse_dict[id_1])
                        for m_id in dup_dict_multiples[reverse_dict[id_1]]:
                            if m_id==id_1: continue
                            if m_id==id_2: continue
                            print('......'*10)



                            r  = df[df['datasetId']==m_id].iloc[0]  # Get first row as Series
                            mm = df[df['datasetId']==m_id].index.values[0]



                            print(f"     - {'Dataset ID':20s}: {r['datasetId']}")
                            print(f"     - {'URL':20s}: {r['originalDataURL']}")


                            # Prepare the text info
                            info_text = (f"{r['datasetId']} \n {r['geo_siteName']} ({r['geo_meanLat']:.1f}, {r['geo_meanLon']:.1f}, {r['geo_meanElev']:.1f}) | "
                                         f"{r['archiveType']} | {r['paleoData_proxy']} | {r['paleoData_variableName']}")

                            # Check for previous duplicates
                            prev_dup_text = ""
                            if len(dup_dec) > 1:
                                if str(mm) in np.array(dup_dec)[:, 0]:
                                    mm_loc = np.array(dup_dec)[:, 0] == str(mm)
                                    prev_dup_text = f" \n DECISION: {np.array(dup_dec)[mm_loc, -4][0]} (detected as pot dup of: {np.array(dup_dec)[mm_loc, 4][0]})"
                                elif str(mm) in np.array(dup_dec)[:, 1]:
                                    mm_loc = np.array(dup_dec)[:, 1] == str(mm)
                                    prev_dup_text = f" \n DECISION: {np.array(dup_dec)[mm_loc, -3][0]}  (detected as pot dup of: {np.array(dup_dec)[mm_loc, 3][0]})"


                            def line_11(x,y):
                                if (len(x)>0) & (len(y)>0):
                                    min_val = min(x.min(), y.min())
                                    max_val = max(x.max(), y.max())
                                    plt.plot([min_val, max_val], [min_val, max_val],color='tab:grey', alpha=0.75, zorder=0, label='1:1 line')
                                return

                            # Create figure with info as title
                            mt, mi1, mi2 = np.intersect1d(df['year'].loc[ii], df['year'].loc[mm], return_indices=True) 


                            multiple_fig = plt.figure(figsize=(4*2, 2), dpi=120)
                            multiple_fig.suptitle(info_text + prev_dup_text, fontsize=8, y=0.98)




                            # First subplot: 
                            plt.subplot(141)
                            x = np.array(df['paleoData_values'].loc[mm])[mi2]
                            y = np.array(df['paleoData_values'].loc[ii])[mi1]
                            plt.scatter(x, y, s=5)
                            plt.ylabel(f'{m_id[:20]}\n{m_id[20:40]}\n{m_id[40:]}'.replace('\n\n',''), fontsize=8)
                            plt.xlabel(f'{id_1[:20]}\n{id_1[20:40]}\n{id_1[40:]}'.replace('\n\n',''), fontsize=8)
                            plt.tick_params(labelsize=7)
                            line_11(x,y)

                            # Second subplot
                            mt, mi1, mi2 = np.intersect1d(df['year'].loc[jj], df['year'].loc[mm], return_indices=True) 
                            plt.subplot(142)
                            x = np.array(df['paleoData_values'].loc[mm])[mi2]
                            y = np.array(df['paleoData_values'].loc[jj])[mi1]
                            plt.scatter(x, y, s=5)
                            plt.ylabel(f'{m_id[:20]}\n{m_id[20:40]}\n{m_id[40:]}'.replace('\n\n',''), fontsize=8)
                            plt.xlabel(f'{id_2[:20]}\n{id_2[20:40]}\n{id_2[40:]}'.replace('\n\n',''), fontsize=8)
                            plt.tick_params(labelsize=7)
                            line_11(x,y)


                            # Third subplot
                            plt.subplot(143)
                            plt.scatter(df['year'].loc[ii], df['paleoData_values'].loc[ii], label=df['datasetId'].loc[ii],
                                       edgecolor='tab:blue', marker='o', s=6, facecolor='None', lw=0.7)
                            plt.scatter(df['year'].loc[mm], df['paleoData_values'].loc[mm], label=df['datasetId'].loc[mm], 
                                        color='k', marker='o', zorder=0, alpha=0.9, s=3)
                            plt.ylabel('year', fontsize=8)
                            plt.xlabel('paleoData_values', fontsize=8)
                            plt.tick_params(labelsize=7)
                            plt.legend(fontsize=8)

                            # Fourth subplot
                            plt.subplot(144)
                            plt.scatter(df['year'].loc[jj], df['paleoData_values'].loc[jj], label=df['datasetId'].loc[jj],
                                       color='tab:red', marker='x', s=6,lw=.7)
                            plt.scatter(df['year'].loc[mm], df['paleoData_values'].loc[mm], label=df['datasetId'].loc[mm], 
                                       color='k', marker='o', zorder=0, alpha=0.9, s=3)
                            plt.ylabel('year', fontsize=8)
                            plt.xlabel('paleoData_values', fontsize=8)
                            plt.tick_params(labelsize=7)
                            plt.legend(fontsize=8)


                            plt.tight_layout()  # Leave room for suptitle
                            plt.show(block=True)

                            save_fig(multiple_fig, '%03d_%s_%s'%(i_pot_dups, id_1, id_2)+f'__{ii}_{jj}_MULTIPLE_{mm}', 
                                     dir=f'/dup_detection/{df.name}', figformat='pdf')



                        print('-------'*15)

                    if comment:
                        print('Before inputting your decision. Would you like to leave a comment on your decision process?')
                        dec_comment = input(f'{i_pot_dups+1}/{n_pot_dups}: **COMMENT** Please type your comment here and/or press enter.')
                    else:
                        dec_comment = ''
                    keep = input(f'{i_pot_dups+1}/{n_pot_dups}: **DECISION** Keep record 1 (%s, blue circles) [1], record 2 (%s, red crosses) [2], keep both [b], keep none [n] or create a composite of both records [c]? Note: only overlapping timesteps are being composited. [Type 1/2/b/n/c]:'%(id_1, id_2))




                    save_fig(fig, '%03d_%s_%s'%(i_pot_dups, id_1, id_2)+f'__{ii}_{jj}', dir=f'/dup_detection/{df.name}', figformat='pdf')


                    figpath    = 'https://nzero.umd.edu:444/hub/user-redirect/lab/tree/dod2k_v2.0/figs/dup_detection/%s'%df.name
                    figpath  += '%03d_%s_%s'%(i_pot_dups, id_1, id_2)+f'__{ii}_{jj}.jpg'

                # now write down the decision
                if keep=='1':
                    print('KEEP BLUE CIRCLES: keep %s, remove %s.'%(id_1, id_2))
                    decisions[ii]='KEEP'
                    decisions[jj]='REMOVE'
                elif keep=='2':
                    print('KEEP RED CROSSES: remove %s, keep %s.'%(id_1, id_2))
                    decisions[jj]='KEEP'
                    decisions[ii]='REMOVE'
                elif keep=='n':
                    print('REMOVE BOTH: remove %s, remove %s.'%(id_2, id_1))
                    decisions[ii]='REMOVE'
                    decisions[jj]='REMOVE'
                elif keep=='b':
                    print('KEEP BOTH: keep %s, keep %s.'%(id_1, id_2))
                    decisions[ii]='KEEP'
                    decisions[jj]='KEEP'
                elif keep=='c':
                    print('CREATE A COMPOSITE OF BOTH RECORDS: %s, %s.'%(id_1, id_2))
                    decisions[ii]='COMPOSITE'
                    decisions[jj]='COMPOSITE'



            cand_pair =[[ii, jj, figpath,
                        id_1, id_2, or_db_1, or_db_2, site_1, site_2,
                        lat_1, lat_2, lon_1, lon_2, 
                        df['geo_meanElev'].loc[ii], df['geo_meanElev'].loc[jj],
                        df['archiveType'].loc[ii], df['archiveType'].loc[jj],
                        df['paleoData_proxy'].loc[ii], df['paleoData_proxy'].loc[jj],
                        df['originalDataURL'].loc[ii], df['originalDataURL'].loc[jj],
                        '%3.1f-%3.1f'%(np.min(df['year'].loc[ii]), np.max(df['year'].loc[ii])), 
                        '%3.1f-%3.1f'%(np.min(df['year'].loc[jj]), np.max(df['year'].loc[jj])), 
                        decisions[ii], decisions[jj],
                        dec_type, dec_comment]]

            dup_dec  += cand_pair
            writer.writerows(cand_pair)
            print('write decision to backup file')
            # raise Exception

    print('=====================================================================')
    print('END OF DUPLICATE DECISION PROCESS.')
    print('=====================================================================')

    comment = '# '+input('Type your comment on your decision process here and/or press enter:')
    header  += [[comment]]
    # header  += cols

    print(np.array(dup_dec).shape)
    filename +=f'_{date_time[2:10]}'
    write_csv(np.array(dup_dec), dirname+filename, header=header, cols=cols)
    print('Saved the decisions under %s.csv'%(dirname+filename))

    print('Summary of all decisions made:')
    for ii_d in range(len(dup_dec)):
        keep_1=dup_dec[ii_d][-4]
        keep_2=dup_dec[ii_d][-3]
        print('#%d: %s record %s. %s record %s.'%(ii_d, keep_1, dup_dec[ii_d][3], 
                                                  keep_2, dup_dec[ii_d][4]))
    return

find_duplicates(df, dist_tolerance_km=8, n_points_thresh=10, corr_thresh=0.9, rmse_thresh=0.1, corr_diff_thresh=0.9, rmse_diff_thresh=0.1, elev_tolerance=0, ignore_same_database=False, save=True, print_output=False)

Identify potential duplicate records in a dataset using metadata and time series similarity.

This is a simpler version of find_duplicates_optimized.

Parameters:

Name Type Description Default
df DataFrame

DataFrame containing proxy records with metadata. Expected columns include: ['geo_meanLat', 'geo_meanLon', 'geo_meanElev', 'year', 'paleoData_values', 'archiveType', 'paleoData_proxy', 'geo_siteName', 'originalDatabase', 'originalDataURL', 'datasetId'].

 required
dist_tolerance_km float

Maximum allowed geographical distance between duplicates (km). Default is 8.

 8
n_points_thresh int

Minimum number of overlapping time points required. Default is 10.

 10
corr_thresh float

Minimum correlation threshold for duplicate detection. Default is 0.9.

 0.9
rmse_thresh float

Maximum RMSE allowed for duplicate detection. Default is 0.1.

 0.1
corr_diff_thresh float

Minimum correlation of first differences threshold. Default is 0.9.

 0.9
rmse_diff_thresh float

Maximum RMSE of first differences allowed. Default is 0.1.

 0.1
elev_tolerance float

Maximum allowed elevation difference. Default is 0.

 0
ignore_same_database bool

If True, ignores potential duplicates within the same database. Default is False.

 False
save bool

If True, saves results as a CSV file. Default is True.

 True
print_output bool

If True, prints progress and matching information. Default is False.

 False

Returns:

Name Type Description
pot_dup_inds list of list of int

Indices of detected potential duplicates.
pot_dup_IDs list of list of str

Dataset IDs of detected potential duplicates.
distances_km ndarray

Matrix of pairwise distances between records.
correlations ndarray

Matrix of pairwise correlations between records.
Source code in dod2k_utilities/ut_duplicate_search.py 
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
def find_duplicates(df, dist_tolerance_km=8, n_points_thresh=10, 
                    corr_thresh=0.9, rmse_thresh=0.1, corr_diff_thresh=0.9,
                    rmse_diff_thresh=0.1, elev_tolerance=0, ignore_same_database=False, 
                    save=True, print_output=False
                   ):
    """
    Identify potential duplicate records in a dataset using metadata and time series similarity.

    This is a simpler version of `find_duplicates_optimized`.

    Parameters
    ----------
    df : pandas.DataFrame
        DataFrame containing proxy records with metadata. Expected columns include:
        ['geo_meanLat', 'geo_meanLon', 'geo_meanElev', 'year', 'paleoData_values', 
         'archiveType', 'paleoData_proxy', 'geo_siteName', 'originalDatabase', 
         'originalDataURL', 'datasetId'].
    dist_tolerance_km : float, optional
        Maximum allowed geographical distance between duplicates (km). Default is 8.
    n_points_thresh : int, optional
        Minimum number of overlapping time points required. Default is 10.
    corr_thresh : float, optional
        Minimum correlation threshold for duplicate detection. Default is 0.9.
    rmse_thresh : float, optional
        Maximum RMSE allowed for duplicate detection. Default is 0.1.
    corr_diff_thresh : float, optional
        Minimum correlation of first differences threshold. Default is 0.9.
    rmse_diff_thresh : float, optional
        Maximum RMSE of first differences allowed. Default is 0.1.
    elev_tolerance : float, optional
        Maximum allowed elevation difference. Default is 0.
    ignore_same_database : bool, optional
        If True, ignores potential duplicates within the same database. Default is False.
    save : bool, optional
        If True, saves results as a CSV file. Default is True.
    print_output : bool, optional
        If True, prints progress and matching information. Default is False.

    Returns
    -------
    pot_dup_inds : list of list of int
        Indices of detected potential duplicates.
    pot_dup_IDs : list of list of str
        Dataset IDs of detected potential duplicates.
    distances_km : numpy.ndarray
        Matrix of pairwise distances between records.
    correlations : numpy.ndarray
        Matrix of pairwise correlations between records.
    """
    n_proxies = len(df)
    print(df.name)

    #database_name = '_'.join(df.originalDatabase.unique())

    # Loop through the proxies
    distances_km    = np.zeros((n_proxies,n_proxies))
    correlations    = np.zeros((n_proxies,n_proxies))
    rmse            = np.zeros((n_proxies,n_proxies))
    corr_diff       = np.zeros((n_proxies,n_proxies))
    rmse_diff       = np.zeros((n_proxies,n_proxies))

    n_pot_dups     = 0 # counts number of detected duplicates
    pot_dup_inds   = [] # stores potential duplicate indices 
    pot_dup_IDs    = [] # stores potential duplicate IDs
    pot_dup_corrs  = [] # stores potential duplicate correlations 
    pot_dup_dists  = [] # stores potential duplicate distances
    pot_dup_meta   = [['index 1', 'index 2', 'ID 1', 'ID 2', 'correlation', 'distance (km)']]

    print('Start duplicate search:')
    print('='*33)
    print('checking parameters:')
    print('%-16s'%'proxy archive                ', ' : %-16s'%' must match')
    print('%-16s'%'proxy type                   ', ' : %-16s'%' must match')
    print('%-16s'%'distance (km)                ', ' < %-16s'%str(dist_tolerance_km)) 
    print('%-16s'%'elevation                    ', ' : %-16s'%' must match')
    print('%-16s'%'time overlap                 ', ' > %-16s'%str(n_points_thresh))
    print('%-16s'%'correlation                  ', ' > %-16s'%str(corr_thresh))
    print('%-16s'%'RMSE                         ', ' < %-16s'%str(rmse_thresh))
    print('%-16s'%'1st difference rmse          ', ' < %-16s'%str(rmse_diff_thresh))
    print('%-16s'%'correlation of 1st difference', ' > %-16s'%str(corr_diff_thresh))
    print('='*33)

    ddir = 'data/%s/dup_detection/'%(df.name)
    fn   = 'dup_detection_candidates_'+df.name
    # fn   = 'pot_dup_%s_'+df.name

    print('Start duplicate search')
    # otherwise re-generate from scratch
    for ii in range(n_proxies):
        if ii in range(0, 10000, 10):
            print('Progress: '+str(ii)+'/'+str(n_proxies))

        # Location of proxy 1
        lat_1   = df['geo_meanLat'].iloc[ii]
        lon_1   = df['geo_meanLon'].iloc[ii]
        # time and data of proxy 1
        time_1_ma = df['year'].iloc[ii]
        data_1_ma = np.array(df['paleoData_values'].iloc[ii])
        if type(time_1_ma)==np.ma.core.MaskedArray:
            time_1  = time_1_ma.data[~data_1_ma.mask]
            data_1  = data_1_ma.data[~data_1_ma.mask]
        else:
            time_1  = time_1_ma
            data_1  = data_1_ma
        # archive and proxy type of proxy 1
        arch_1  = df['archiveType'].iloc[ii].lower()
        type_1  = df['paleoData_proxy'].iloc[ii].lower()
        site_1  = df['geo_siteName'].iloc[ii].lower()
        db_1    = df['originalDatabase'].iloc[ii].lower()
        url_1   = df['originalDataURL'].iloc[ii].lower()
        id_1    = df['datasetId'].iloc[ii].lower()
        #
        for jj in range(ii+1, n_proxies):
            db_2 = df['originalDatabase'].iloc[jj].lower()
            id_2    = df['datasetId'].iloc[jj].lower()

            if (db_1==db_2) & ignore_same_database: continue
            # Location of proxy 2
            lat_2 = df['geo_meanLat'].iloc[jj]
            lon_2 = df['geo_meanLon'].iloc[jj]
            # time and data of proxy 2
            time_2_ma = df['year'].iloc[jj]
            data_2_ma = np.array(df['paleoData_values'].iloc[jj])

            # print(jj, db_2, type(time_2_ma), type(data_2_ma))

            if type(time_2_ma)==np.ma.core.MaskedArray:
                time_2  = time_2_ma.data[~data_2_ma.mask]
                data_2  = data_2_ma.data[~data_2_ma.mask]
            else:
                time_2  = time_2_ma
                data_2  = data_2_ma

            # archive and proxy type of proxy 2
            arch_2  = df['archiveType'].iloc[jj].lower()
            type_2  = df['paleoData_proxy'].iloc[jj].lower()
            site_2  = df['geo_siteName'].iloc[jj].lower()
            db_2    = df['originalDatabase'].iloc[jj].lower()
            url_2   = df['originalDataURL'].iloc[jj].lower()

            # Calculate distance between proxy 1 and proxy 2 in km
            distances_km[ii, jj] = geopy.distance.great_circle((lat_1,lon_1),
                                                               (lat_2,lon_2)).km
            # find shared time values 
            time_12, int_1, int_2 = np.intersect1d(time_1, time_2, return_indices=True) # possibly need to round these time series?

            if time_12.shape[0]<=n_points_thresh: 
                #print('%d|%d no shared time'%(ii,jj))
                continue

            # calculate correlation, rms, sum of differences between shared i and j data
            # z_1 is the first difference between each datapoints for record 1
            z_1  = (data_1[int_1][:-1]-data_1[int_1][1:])
            z_1 -= np.mean(z_1)        
            if np.std(z_1)!=0: 
                z_1 /= np.std(z_1)


            # z_2 is the first difference between each datapoints for record 2
            z_2  = (data_2[int_2][:-1]-data_2[int_2][1:])
            z_2 -= np.mean(z_2)
            if np.std(z_2)!=0: 
                z_2 /= np.std(z_2)


            correlations[ii, jj] = np.corrcoef(np.vstack([data_1[int_1], data_2[int_2]]))[0,1]

            rmse[ii, jj]         = np.sqrt(np.sum((data_1[int_1]-data_2[int_2])**2)/len(time_12))


            rmse_diff[ii, jj]    = np.sqrt(np.sum((z_1-z_2)**2)/len(time_12))
            corr_diff[ii, jj]    = np.corrcoef(np.vstack([z_1, z_2]))[0,1]



            if ((np.isnan(correlations[ii,jj]) & np.isnan(rmse[ii,jj]))|(np.isnan(corr_diff[ii,jj]) & np.isnan(rmse_diff[ii,jj]))):
                print('!!! %d|%d nan detected in both correlation and rmse of record data (or its 1st difference). Danger of missing duplicate!!'%(ii, jj))
                for idd, dd in enumerate([correlations[ii,jj], rmse[ii,jj], rmse_diff[ii,jj],  corr_diff[ii,jj]]):
                    if np.isnan(dd):
                        print('nan detected in %s'%(['correlation','RMSE','RMSE of 1st difference', 'correlation of 1st difference'][idd]))

            # DEFINE CRITERIA:

            meta_crit         = (arch_1 == arch_2) & (type_1 == type_2) # archive types and proxy types must agree
            elevation_not_nan = ((~np.isnan(df['geo_meanElev'].iloc[ii]))& ~np.isnan(df['geo_meanElev'].iloc[jj]))
            elevation_dist    = np.abs(df['geo_meanElev'].iloc[ii]-df['geo_meanElev'].iloc[jj])


            location_crit = ((distances_km[ii,jj] <= dist_tolerance_km) & 
                             (elevation_dist<=elev_tolerance if elevation_not_nan else True))
            #print('loc', location_crit)
    # i,j are located within dist_tolerance_km of each others and have the same elevation (provided they're not nans)
            overlap_crit  = ((len(time_12) > n_points_thresh) |
                             (len(time_1) <= n_points_thresh) |
                             (len(time_2)  <= n_points_thresh) )  # we have a sufficient time overlap between the datasets 
            site_crit     = (np.any([s1 in s2 for s2 in site_2.split(' ') for s1 in site_1.split(' ')])| 
                             np.any([s1 in s2 for s2 in site_2.split('-') for s1 in site_1.split('-')])| 
                             np.any([s1 in s2 for s2 in site_2.split('_') for s1 in site_1.split('_')])|
                             np.any([site_1 in site_2, site_2 in site_1])) # there is at least some overlap in the site name 
            corr_crit     = (((correlations[ii, jj] > corr_thresh) | 
                             (corr_diff[ii, jj] > corr_diff_thresh)) # correlation between shared datapoints exceeds correlation threshol dor correlation of first difference above threshold
                             & 
                             ((rmse[ii, jj] < rmse_thresh) | 
                             (rmse_diff[ii, jj] < rmse_diff_thresh))) #  RMSE of records or of first difference below threshold
            url_crit       = (url_1==url_2 if db_1==db_2 else True)
            if print_output:
                print('--------')
                print(id_1, id_2)
                print('archive and proxy match: %s'%meta_crit)
                print('site match: %s'%site_crit)
                print('location match: %s'%location_crit)
                print('correlation high: %s'%corr_crit)
                print('overlap %s'%overlap_crit)
            if (meta_crit & site_crit #& url_crit
                & location_crit
                & corr_crit
                & overlap_crit)|((correlations[ii, jj] > 0.98)& overlap_crit):
                # if these criteria are satisfied, we found a possible duplicate!
                n_pot_dups    += 1
                pot_dup_inds  += [[ii, jj]]
                pot_dup_IDs   += [df['datasetId'].iloc[[ii, jj]].values]
                pot_dup_corrs += [[correlations[ii, jj]]]
                pot_dup_dists += [[distances_km[ii,jj]]]
                pot_dup_meta  += [[ii, jj, df['datasetId'].iloc[ii], df['datasetId'].iloc[jj], correlations[ii, jj], distances_km[ii,jj]]]
                print('--> Found potential duplicate: %d: %s&%d: %s (n_potential_duplicates=%d)'%(ii, id_1, jj, id_2, n_pot_dups))

    if save: 
        os.makedirs(ddir, exist_ok=True)
        write_csv(pot_dup_meta, ddir+fn)

    print('='*60)
    print('Saved indices, IDs, distances, correlations in %s'%ddir)
    print('='*60)
    print('Detected %d possible duplicates in %s.'%(n_pot_dups, df.name))
    print('='*60)
    if len(pot_dup_inds)<=10:
        print('='*60)
        print('Indices: %s'%(', ').join([str(pdi) for pdi in pot_dup_inds]))
        print('IDs: %s'%(', ').join([pdi[0]+' + '+pdi[1] for pdi in pot_dup_IDs]))
        print('='*60)
    return #pot_dup_inds, pot_dup_IDs, distances_km, correlations

find_duplicates_optimized(df, dist_tolerance_km=8, n_points_thresh=10, corr_thresh=0.9, rmse_thresh=0.1, corr_diff_thresh=0.9, rmse_diff_thresh=0.1, elev_tolerance=0, ignore_same_database=False, save=True, print_output=False, return_data=False)

Identify potential duplicate records in a dataset based on metadata and time series similarity.

Based on find_duplicates, optimized by Feng Zhu

Parameters:

Name Type Description Default
df DataFrame

DataFrame containing proxy records with metadata. Expected columns include: ['geo_meanLat', 'geo_meanLon', 'geo_meanElev', 'year', 'paleoData_values', 'archiveType', 'paleoData_proxy', 'geo_siteName', 'originalDatabase', 'originalDataURL', 'datasetId'].

 required
dist_tolerance_km float

Maximum allowed geographical distance between duplicates (km). Default is 8.

 8
n_points_thresh int

Minimum number of overlapping time points required. Default is 10.

 10
corr_thresh float

Minimum correlation threshold for duplicate detection. Default is 0.9.

 0.9
rmse_thresh float

Maximum RMSE allowed for duplicate detection. Default is 0.1.

 0.1
corr_diff_thresh float

Minimum correlation of first differences threshold. Default is 0.9.

 0.9
rmse_diff_thresh float

Maximum RMSE of first differences allowed. Default is 0.1.

 0.1
elev_tolerance float

Maximum allowed elevation difference. Default is 0.

 0
ignore_same_database bool

If True, ignores potential duplicates within the same database. Default is False.

 False
save bool

If True, saves results as a CSV file. Default is True.

 True
print_output bool

If True, prints progress and matching information. Default is False.

 False

Returns:

Name Type Description
pot_dup_inds list of list of int

Indices of detected potential duplicates.
pot_dup_IDs list of list of str

Dataset IDs of detected potential duplicates.
distances_km ndarray

Matrix of pairwise distances between records.
correlations ndarray

Matrix of pairwise correlations between records.
Source code in dod2k_utilities/ut_duplicate_search.py 
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
def find_duplicates_optimized(df, dist_tolerance_km=8, n_points_thresh=10, 
                    corr_thresh=0.9, rmse_thresh=0.1, corr_diff_thresh=0.9,
                    rmse_diff_thresh=0.1, elev_tolerance=0, ignore_same_database=False, 
                    save=True, print_output=False, return_data=False
                   ):
    """
    Identify potential duplicate records in a dataset based on metadata and time series similarity.

    Based on find_duplicates, optimized by Feng Zhu

    Parameters
    ----------
    df : pandas.DataFrame
        DataFrame containing proxy records with metadata. Expected columns include:
        ['geo_meanLat', 'geo_meanLon', 'geo_meanElev', 'year', 'paleoData_values', 
         'archiveType', 'paleoData_proxy', 'geo_siteName', 'originalDatabase', 
         'originalDataURL', 'datasetId'].
    dist_tolerance_km : float, optional
        Maximum allowed geographical distance between duplicates (km). Default is 8.
    n_points_thresh : int, optional
        Minimum number of overlapping time points required. Default is 10.
    corr_thresh : float, optional
        Minimum correlation threshold for duplicate detection. Default is 0.9.
    rmse_thresh : float, optional
        Maximum RMSE allowed for duplicate detection. Default is 0.1.
    corr_diff_thresh : float, optional
        Minimum correlation of first differences threshold. Default is 0.9.
    rmse_diff_thresh : float, optional
        Maximum RMSE of first differences allowed. Default is 0.1.
    elev_tolerance : float, optional
        Maximum allowed elevation difference. Default is 0.
    ignore_same_database : bool, optional
        If True, ignores potential duplicates within the same database. Default is False.
    save : bool, optional
        If True, saves results as a CSV file. Default is True.
    print_output : bool, optional
        If True, prints progress and matching information. Default is False.

    Returns
    -------
    pot_dup_inds : list of list of int
        Indices of detected potential duplicates.
    pot_dup_IDs : list of list of str
        Dataset IDs of detected potential duplicates.
    distances_km : numpy.ndarray
        Matrix of pairwise distances between records.
    correlations : numpy.ndarray
        Matrix of pairwise correlations between records.
    """
    n_proxies = len(df)
    print(df.name)

    #database_name = '_'.join(df.originalDatabase.unique())

    # Loop through the proxies
    distances_km    = np.zeros((n_proxies,n_proxies))
    correlations    = np.zeros((n_proxies,n_proxies))
    rmse            = np.zeros((n_proxies,n_proxies))
    corr_diff       = np.zeros((n_proxies,n_proxies))
    rmse_diff       = np.zeros((n_proxies,n_proxies))

    n_pot_dups     = 0 # counts number of detected duplicates
    pot_dup_inds   = [] # stores potential duplicate indices 
    pot_dup_IDs    = [] # stores potential duplicate IDs
    pot_dup_corrs  = [] # stores potential duplicate correlations 
    pot_dup_dists  = [] # stores potential duplicate distances
    pot_dup_meta   = [['index 1', 'index 2', 'ID 1', 'ID 2', 'correlation', 'distance (km)']]

    print('Start duplicate search:')
    print('='*33)
    print('checking parameters:')
    print('%-16s'%'proxy archive                ', ' : %-16s'%' must match')
    print('%-16s'%'proxy type                   ', ' : %-16s'%' must match')
    print('%-16s'%'distance (km)                ', ' < %-16s'%str(dist_tolerance_km)) 
    print('%-16s'%'elevation                    ', ' : %-16s'%' must match')
    print('%-16s'%'time overlap                 ', ' > %-16s'%str(n_points_thresh))
    print('%-16s'%'correlation                  ', ' > %-16s'%str(corr_thresh))
    print('%-16s'%'RMSE                         ', ' < %-16s'%str(rmse_thresh))
    print('%-16s'%'1st difference rmse          ', ' < %-16s'%str(rmse_diff_thresh))
    print('%-16s'%'correlation of 1st difference', ' > %-16s'%str(corr_diff_thresh))
    print('='*33)

    ddir = 'data/%s/dup_detection/'%(df.name)
    fn   = 'dup_detection_candidates_'+df.name


    print('Start duplicate search')
    # otherwise re-generate from scratch
    for ii in range(n_proxies):
        if ii in range(0, 10000, 10):
            print('Progress: '+str(ii)+'/'+str(n_proxies))

        # Location of proxy 1
        lat_1   = df['geo_meanLat'].iloc[ii]
        lon_1   = df['geo_meanLon'].iloc[ii]
        # time and data of proxy 1
        time_1_ma = df['year'].iloc[ii]
        data_1_ma = np.array(df['paleoData_values'].iloc[ii])
        if type(time_1_ma)==np.ma.core.MaskedArray:
            time_1  = time_1_ma.data[~data_1_ma.mask]
            data_1  = data_1_ma.data[~data_1_ma.mask]
        else:
            time_1  = time_1_ma
            data_1  = data_1_ma
        # archive and proxy type of proxy 1
        arch_1  = df['archiveType'].iloc[ii].lower()
        type_1  = df['paleoData_proxy'].iloc[ii].lower()
        site_1  = df['geo_siteName'].iloc[ii].lower()
        db_1    = df['originalDatabase'].iloc[ii].lower()
        url_1   = df['originalDataURL'].iloc[ii].lower()
        id_1    = df['datasetId'].iloc[ii].lower()
        #
        for jj in range(ii+1, n_proxies):
            db_2 = df['originalDatabase'].iloc[jj].lower()
            id_2    = df['datasetId'].iloc[jj].lower()

            if (db_1==db_2) & ignore_same_database: continue
            # Location of proxy 2
            lat_2 = df['geo_meanLat'].iloc[jj]
            lon_2 = df['geo_meanLon'].iloc[jj]
            # time and data of proxy 2
            time_2_ma = df['year'].iloc[jj]
            data_2_ma = np.array(df['paleoData_values'].iloc[jj])


            if type(time_2_ma)==np.ma.core.MaskedArray:
                time_2  = time_2_ma.data[~data_2_ma.mask]
                data_2  = data_2_ma.data[~data_2_ma.mask]
            else:
                time_2  = time_2_ma
                data_2  = data_2_ma

            # archive and proxy type of proxy 2
            arch_2  = df['archiveType'].iloc[jj].lower()
            type_2  = df['paleoData_proxy'].iloc[jj].lower()
            site_2  = df['geo_siteName'].iloc[jj].lower()
            db_2    = df['originalDatabase'].iloc[jj].lower()
            url_2   = df['originalDataURL'].iloc[jj].lower()

            ### Criterion on matching metadata
            meta_crit         = (arch_1 == arch_2) & (type_1 == type_2) # archive types and proxy types must agree
            if not meta_crit: continue

            ### Criterion on elevation
            elevation_not_nan = ((~np.isnan(df['geo_meanElev'].iloc[ii]))& ~np.isnan(df['geo_meanElev'].iloc[jj]))
            elevation_dist    = np.abs(df['geo_meanElev'].iloc[ii]-df['geo_meanElev'].iloc[jj])
            if not elevation_not_nan and elevation_dist>elev_tolerance: continue

            ### Criterion on time overlap
            time_12, int_1, int_2 = np.intersect1d(time_1, time_2, return_indices=True) # possibly need to round these time series?
            if time_12.shape[0]<=n_points_thresh: continue

            ### Criterion on distance
            # Calculate distance between proxy 1 and proxy 2 in km
            distances_km[ii, jj] = geopy.distance.great_circle((lat_1,lon_1), (lat_2,lon_2)).km
            if distances_km[ii,jj] > dist_tolerance_km: continue

            # calculate correlation, rms, sum of differences between shared i and j data
            z_1  = (data_1[int_1][:-1]-data_1[int_1][1:])
            z_1 -= np.mean(z_1)        
            if np.std(z_1)!=0: 
                z_1 /= np.std(z_1)
            # else:
            #     plt.figure()
            #     plt.scatter(time_1, data_1, label='data_1 (%s)'%id_1)
            #     plt.scatter(time_12, data_1[int_1], label='data_1 SHARED TIME (%s)'%id_1)
            #     plt.scatter(time_2, data_2, label='data_2 (%s)'%id_2)
            #     plt.scatter(time_12, data_2[int_2], label='data_2 SHARED TIME (%s)'%id_2)

            #     plt.legend()
            #     plt.show()


            z_2  = (data_2[int_2][:-1]-data_2[int_2][1:])
            z_2 -= np.mean(z_2)
            if np.std(z_2)!=0: 
                z_2 /= np.std(z_2)
            # else:

            #     plt.figure()

            #     plt.scatter(time_12, data_1[int_1]-np.mean(data_1[int_1]), label='data_1 SHARED TIME (%s)'%id_1)

            #     plt.scatter(time_12, data_2[int_2]-np.mean(data_2[int_2]), label='data_2 SHARED TIME (%s)'%id_2)

            #     plt.legend()
            #     plt.show()


            correlations[ii, jj] = np.corrcoef(np.vstack([data_1[int_1], data_2[int_2]]))[0,1]

            rmse[ii, jj]         = np.sqrt(np.sum((data_1[int_1]-data_2[int_2])**2)/len(time_12))


            rmse_diff[ii, jj]    = np.sqrt(np.sum((z_1-z_2)**2)/len(time_12))
            corr_diff[ii, jj]    = np.corrcoef(np.vstack([z_1, z_2]))[0,1]



            if ((np.isnan(correlations[ii,jj]) & np.isnan(rmse[ii,jj]))|(np.isnan(corr_diff[ii,jj]) & np.isnan(rmse_diff[ii,jj]))):
                print('!!! %d|%d nan detected in both correlation and rmse of record data (or its 1st difference). Danger of missing duplicate!!'%(ii, jj))
                for idd, dd in enumerate([correlations[ii,jj], rmse[ii,jj], rmse_diff[ii,jj],  corr_diff[ii,jj]]):
                    if np.isnan(dd):
                        print('nan detected in %s'%(['correlation','RMSE','RMSE of 1st difference', 'correlation of 1st difference'][idd]))

            # DEFINE CRITERIA:


    # i,j are located within dist_tolerance_km of each others and have the same elevation (provided they're not nans)
            overlap_crit  = ((len(time_12) > n_points_thresh) |
                             (len(time_1) <= n_points_thresh) |
                             (len(time_2)  <= n_points_thresh) )  # we have a sufficient time overlap between the datasets 
            site_crit     = np.any([s1 in s2 for s2 in site_2.split(' ') for s1 in site_1.split(' ')])| np.any([s1 in s2 for s2 in site_2.split('-') for s1 in site_1.split('-')])| np.any([s1 in s2 for s2 in site_2.split('_') for s1 in site_1.split('_')])|np.any([site_1 in site_2, site_2 in site_1]) # there is at least some overlap in the site name 
            corr_crit     = (((correlations[ii, jj] > corr_thresh) | 
                             (corr_diff[ii, jj] > corr_diff_thresh)) # correlation between shared datapoints exceeds correlation threshol dor correlation of first difference above threshold
                             & 
                             ((rmse[ii, jj] < rmse_thresh) | 
                             (rmse_diff[ii, jj] < rmse_diff_thresh))) #  RMSE of records or of first difference below threshold
            url_crit       = (url_1==url_2 if db_1==db_2 else True)
            if print_output:
                print('--------')
                print(id_1, id_2)
                print('archive and proxy match: %s'%meta_crit)
                print('site match: %s'%site_crit)
                # print('location match: %s'%location_crit)
                print('correlation high: %s'%corr_crit)
                print('overlap %s'%overlap_crit)
            if (meta_crit & site_crit #& url_crit
                # & location_crit
                & corr_crit
                & overlap_crit)|((correlations[ii, jj] > 0.98)& overlap_crit):
                # if these criteria are satisfied, we found a possible duplicate!
                n_pot_dups    += 1
                pot_dup_inds  += [[ii, jj]]
                pot_dup_IDs   += [df['datasetId'].iloc[[ii, jj]].values]
                pot_dup_corrs += [[correlations[ii, jj]]]
                pot_dup_dists += [[distances_km[ii,jj]]]
                pot_dup_meta  += [[ii, jj, df['datasetId'].iloc[ii], df['datasetId'].iloc[jj], correlations[ii, jj], distances_km[ii,jj]]]
                print('--> Found potential duplicate: %d: %s&%d: %s (n_potential_duplicates=%d)'%(ii, id_1, jj, id_2, n_pot_dups))

    if save: 
        os.makedirs(ddir, exist_ok=True)
        write_csv(pot_dup_meta, ddir+fn)

        print('='*60)
        print('Saved indices, IDs, distances, correlations in %s'%ddir)
    print('='*60)
    print('Detected %d possible duplicates in %s.'%(n_pot_dups, df.name))
    print('='*60)
    if len(pot_dup_inds)<=10:
        print('='*60)
        print('Indices: %s'%(', ').join([str(pdi) for pdi in pot_dup_inds]))
        print('IDs: %s'%(', ').join([pdi[0]+' + '+pdi[1] for pdi in pot_dup_IDs]))
        print('='*60)
    if return_data:
        return pot_dup_IDs
    return #pot_dup_inds, pot_dup_IDs, distances_km, correlations

join_composites_metadata(df, comp_ID_pairs, df_decisions, header)

Create composite records from overlapping duplicate proxy datasets and generate metadata.

This function combines pairs of proxy records that were identified as duplicates and decided to be composited. It standardizes the data as z-scores, averages overlapping periods, merges metadata, and generates a composite DataFrame. A scatter plot of the original and composite data is created and saved for visual inspection.

Parameters:

Name Type Description Default
df DataFrame

Original DataFrame containing proxy data and metadata. Must include columns: - 'paleoData_values' : proxy values - 'year' : time vector - 'geo_meanLat', 'geo_meanLon', 'geo_meanElev' : site coordinates - 'archiveType', 'geo_siteName', 'paleoData_proxy' : metadata - 'climateInterpretation_variable', 'dataSetName', 'originalDatabase', 'originalDataURL', 'paleoData_notes', 'duplicateDetails' - 'datasetId' : unique record identifier

 required
comp_ID_pairs DataFrame

DataFrame listing pairs of record IDs to be composited. Must include columns: - 'datasetId 1', 'datasetId 2' - 'originalDatabase 1', 'originalDatabase 2' - 'Decision type', 'Decision comment'

 required
df_decisions DataFrame

DataFrame containing the decisions made during duplicate evaluation. Used to annotate the composite metadata with decision type and comments.

 required
header list

Metadata header from the duplicate decision process. Used for documenting operator details in the composite notes.

 required

Returns:

Name Type Description
df_comp DataFrame

A new DataFrame containing the composited proxy records, including: - Combined 'paleoData_values' as z-scores - Merged 'year' vector - Updated metadata fields, including a composite 'datasetId' - Detailed 'duplicateDetails' recording the composition process
Notes
  • For numerical metadata that differs between records, the mean is taken.
  • For categorical metadata, entries are concatenated into a composite string.
  • Overlapping periods are averaged, and non-overlapping periods are appended.
  • A scatter plot is generated for each composite showing the original records and the resulting composite, and it is saved using save_fig.
  • The function maintains provenance of original datasets, including notes and URLs, in the 'duplicateDetails' field.
Source code in dod2k_utilities/ut_duplicate_search.py 
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
def join_composites_metadata(df, comp_ID_pairs, df_decisions, header):
    """
    Create composite records from overlapping duplicate proxy datasets and generate metadata.

    This function combines pairs of proxy records that were identified as duplicates
    and decided to be composited. It standardizes the data as z-scores, averages overlapping 
    periods, merges metadata, and generates a composite DataFrame. A scatter plot of the 
    original and composite data is created and saved for visual inspection.

    Parameters
    ----------
    df : pd.DataFrame
        Original DataFrame containing proxy data and metadata. Must include columns:
            - 'paleoData_values' : proxy values
            - 'year' : time vector
            - 'geo_meanLat', 'geo_meanLon', 'geo_meanElev' : site coordinates
            - 'archiveType', 'geo_siteName', 'paleoData_proxy' : metadata
            - 'climateInterpretation_variable', 'dataSetName', 'originalDatabase', 'originalDataURL', 'paleoData_notes', 'duplicateDetails'
            - 'datasetId' : unique record identifier

    comp_ID_pairs : pd.DataFrame
        DataFrame listing pairs of record IDs to be composited. Must include columns:
            - 'datasetId 1', 'datasetId 2'
            - 'originalDatabase 1', 'originalDatabase 2'
            - 'Decision type', 'Decision comment'

    df_decisions : pd.DataFrame
        DataFrame containing the decisions made during duplicate evaluation.
        Used to annotate the composite metadata with decision type and comments.

    header : list
        Metadata header from the duplicate decision process. Used for documenting
        operator details in the composite notes.

    Returns
    -------
    df_comp : pd.DataFrame
        A new DataFrame containing the composited proxy records, including:
            - Combined 'paleoData_values' as z-scores
            - Merged 'year' vector
            - Updated metadata fields, including a composite 'datasetId'
            - Detailed 'duplicateDetails' recording the composition process

    Notes
    -----
    - For numerical metadata that differs between records, the mean is taken.
    - For categorical metadata, entries are concatenated into a composite string.
    - Overlapping periods are averaged, and non-overlapping periods are appended.
    - A scatter plot is generated for each composite showing the original records
      and the resulting composite, and it is saved using `save_fig`.
    - The function maintains provenance of original datasets, including notes and URLs,
      in the 'duplicateDetails' field.
    """

    # create composite dataframe
    df_comp = pd.DataFrame()

    for ii in comp_ID_pairs.index:

        # if ii not in df.datasetId.values: 
        #     print(ii)
        #     print(df.datasetId.values)
        #     print(f'Skip {ii} - allowed for testing only!!')
        #     continue

        row = {}
        ID_1, ID_2   = comp_ID_pairs.loc[ii, ['datasetId 1', 'datasetId 2']]
        db_1, db_2   = comp_ID_pairs.loc[ii, ['originalDatabase 1', 'originalDatabase 2']]
        dec_1, dec_2 = comp_ID_pairs.loc[ii, ['Decision 1', 'Decision 2']]
        print(ID_1, ID_2)

        # metadata should match exactly for the following columns (check), if not choose metadata values:
        add_dup_note = ''
        for key in ['archiveType', 'geo_meanElev', 'geo_meanLat', 'geo_meanLon',
                    'geo_siteName', 'paleoData_proxy', 'yearUnits', 'interpretation_variable', 'interpretation_direction', 'interpretation_seasonality']:

            md1, md2 = df.at[ID_1, key], df.at[ID_2, key]

            if isinstance(md1, (float, np.floating, np.float32, np.float64)):
                md1 = np.round(md1, 3)

            if isinstance(md2, (float, np.floating, np.float32, np.float64)):
                md2 = np.round(md2, 3)

            if md1==md2:
                row[key]=md1
                add_dup_note += f'{key}: Metadata identical'
            else:
                print('--------------------------------------------------------------------------------')
                add_dup_note = 'Metadata differs for %s in original records: %s (%s) and %s (%s). '%(key, ID_1, str(df.at[ID_1, key]), 
                                                                                 ID_2, str(df.at[ID_2, key]))
                print('Metadata different for >>>%s<<< in: %s (%s) and %s (%s). '%(key, ID_1, str(df.at[ID_1, key]), 
                                                                                 ID_2, str(df.at[ID_2, key])))
                try:
                    entry='COMPOSITE: '+df.at[ID_1, key]+' + '+df.at[ID_2, key]
                    add_dup_note += f'{key}: Metadata composited to: '+entry
                    if key in ['interpretation_variable', 'interpretation_direction', 'interpretation_seasonality']:
                        if df.at[ID_1, key]=='N/A': entry=df.at[ID_2, key]
                        if df.at[ID_2, key]=='N/A': entry=df.at[ID_1, key]
                        if (df.at[ID_1, key] in df.at[ID_2, key]):  entry=df.at[ID_2, key]
                        if (df.at[ID_2, key] in df.at[ID_1, key]):  entry=df.at[ID_1, key]
                    loop=False
                except TypeError:
                    # print('Can not create composites for numerical metadata! Create average instead.')
                    # av = input('Type [y] if you want to average the metadata. Otherwise type [n].')
                    # if av.lower() in ['y', 'yes']:
                    entry = np.mean([df.at[ID_1, key], df.at[ID_2, key]])                        
                    add_dup_note += f'{key}: Metadata averaged to: '+str(entry)
                    loop=False
                row[key]= entry
            # print('Add the following note to duplicateDetails:', add_dup_note)
        # create new entries for the following columns:
        # create z-scores of data
        data_1, data_2 = np.array(df.at[ID_1, 'paleoData_values']), np.array(df.at[ID_2, 'paleoData_values'])

        # year
        time_1, time_2 = np.array(df.at[ID_1, 'year']), np.array(df.at[ID_2, 'year'])
        time_12, ii_1, ii_2 = np.intersect1d(time_1, time_2, return_indices=True)
        ii_1x   = [ii for ii in range(len(time_1)) if ii not in ii_1]
        ii_2x   = [ii for ii in range(len(time_2)) if ii not in ii_2]

        data_1 /= np.std(data_1[ii_1])
        data_1 -= np.mean(data_1[ii_1])
        data_2 /= np.std(data_2[ii_2])
        data_2 -= np.mean(data_2[ii_2])

        data_12 = (data_1[ii_1]+data_2[ii_2])/2.

        data = list(data_1[ii_1x])+list(data_12)+list(data_2[ii_2x])
        time = list(time_1[ii_1x])+list(time_12)+list(time_2[ii_2x])

        row['paleoData_values'] = data
        row['year'] = time

        fig = plt.figure(figsize=(6, 2), dpi=100)

        plt.scatter(time_1, data_1, s=20, color='tab:blue', label=ID_1)
        plt.scatter(time_2, data_2, s=20, color='tab:orange', label=ID_2)
        plt.scatter(time, data, s=10, color='k', label='composite')
        plt.legend()
        plt.show()
        save_fig(fig, 'composite_%s_%s'%(ID_1, ID_2), dir='/%s/dup_detection/'%df.name)


        # new metadata for identification etc.
        row['dataSetName']      = df.at[ID_1, 'dataSetName']+', '+df.at[ID_2, 'dataSetName']
        row['originalDatabase'] = 'dod2k_composite_z'
        row['originalDataURL']  = '%s: %s, %s: %s'%(ID_1, df.at[ID_1, 'originalDataURL'], ID_2, df.at[ID_2, 'originalDataURL'])
        row['paleoData_notes']  = '%s: %s, %s: %s'%(ID_1, df.at[ID_1, 'paleoData_notes'], ID_2, df.at[ID_2, 'paleoData_notes'])
        row['interpretation_variableDetail']  = '%s: %s, %s: %s'%(ID_1, df.at[ID_1, 'interpretation_variableDetail'], ID_2, df.at[ID_2, 'interpretation_variableDetail'])
        row['datasetId']        = 'dod2k_composite_z_'+ID_1.replace('dod2k_composite_z_','')+'_'+ID_2.replace('dod2k_composite_z_','')   
        row['paleoData_units']  = 'z-scores'


        # save details on duplicates in duplicateDetails:
        row['duplicateDetails']={'duplicate ID': f'{ID_1} and {ID_2}',
                                 'duplicate database': f'{db_1} and {db_2}',
                                 'duplicate decision': 'COMPOSITE',
                                 'decision type': df_decisions.loc[ii, 'Decision type']}


        # save details on composition process in duplicateDetails:
        row['duplicateDetails']['composite details'] = add_dup_note


        # save operator details
        if comp_ID_pairs.at[ii, 'Decision type']=='MANUAL': 
            operator_details = ' '.join(header[1:]).replace(' Modified ','')[:-2].replace(':','').replace('  E-Mail', '')
            row['duplicateDetails']['operator'] = operator_details
            row['duplicateDetails']['note'] = comp_ID_pairs.at[ii, 'Decision comment']



        # migrate original duplicate details

        row['duplicateDetails']['original duplicate details'] = {ID_1: df.at[ID_1, 'duplicateDetails'], ID_2: df.at[ID_2, 'duplicateDetails']}


        # create dataframe for composites
        df_comp = pd.concat([df_comp, pd.DataFrame({kk: [vv] for kk, vv in row.items()})], ignore_index = True, axis=0)

        # print(df_comp)
    return df_comp

plot_duplicates(df, save_figures=True, write_output=True, display=False)

Generate plots of potential duplicate records in a proxy database and optionally save their metadata.

This function identifies potential duplicate entries in a paleoclimate proxy dataset, visualizes them using the dup_plot function, and saves the metadata of duplicates as a CSV.

Parameters:

Name Type Description Default
df DataFrame

The dataframe containing proxy data and metadata. Expected columns include: - 'geo_meanLat', 'geo_meanLon' : Latitude and longitude of the site. - 'year', 'paleoData_values' : Time vector and proxy values. - 'archiveType', 'paleoData_proxy' : Archive type and proxy type. - 'datasetId' : Unique identifier or dataset name. - 'geo_siteName' : Site name. - 'originalDatabase' : Name of the original database.

 required
save_figures bool

If True, save the generated figures to disk. Default is True.

 True
write_output bool

If True, save the duplicate metadata to a CSV file. Default is True.

 True

Returns:

Type Description
None

The function primarily generates plots and writes metadata; it does not return any objects.
Notes
  • The function assumes that potential duplicates have already been identified by a separate process (e.g., find_duplicates) and that a corresponding CSV exists in df.name/dup_detection/.
  • Metadata printed and saved includes site names, coordinates, dataset IDs, original database, and summary statistics (mean, std, units, etc.).
  • Figures are saved in PDF format by default using save_fig.
Source code in dod2k_utilities/ut_duplicate_search.py 
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
def plot_duplicates(df, save_figures=True, write_output=True, display=False):
    """
    Generate plots of potential duplicate records in a proxy database and optionally save their metadata.

    This function identifies potential duplicate entries in a paleoclimate proxy dataset,
    visualizes them using the `dup_plot` function, and saves the metadata of duplicates as a CSV.

    Parameters
    ----------
    df : pd.DataFrame
        The dataframe containing proxy data and metadata. Expected columns include:
            - 'geo_meanLat', 'geo_meanLon' : Latitude and longitude of the site.
            - 'year', 'paleoData_values' : Time vector and proxy values.
            - 'archiveType', 'paleoData_proxy' : Archive type and proxy type.
            - 'datasetId' : Unique identifier or dataset name.
            - 'geo_siteName' : Site name.
            - 'originalDatabase' : Name of the original database.
    save_figures : bool, optional
        If True, save the generated figures to disk. Default is True.
    write_output : bool, optional
        If True, save the duplicate metadata to a CSV file. Default is True.

    Returns
    -------
    None
        The function primarily generates plots and writes metadata; it does not return any objects.

    Notes
    -----
    - The function assumes that potential duplicates have already been identified by a separate
      process (e.g., `find_duplicates`) and that a corresponding CSV exists in `df.name/dup_detection/`.
    - Metadata printed and saved includes site names, coordinates, dataset IDs, original database,
      and summary statistics (mean, std, units, etc.).
    - Figures are saved in PDF format by default using `save_fig`.
    """


    keys_to_print = ['originalDatabase', 'originalDataURL', 'datasetId', 'archiveType', 'proxy | variableName', #'archive|proxy',
                     'geo_siteName', 'lat | lon | elev', 'mean | std | units', 'year' ]

    ddir = 'data/%s/dup_detection/'%(df.name)+'dup_detection_candidates_'+df.name
    # load the potential duplicate data as found in find_duplicates function
    pot_dup_meta, head = read_csv(ddir, header=True)

    pot_dup_inds   = np.array(np.array(pot_dup_meta)[:, :2], dtype=int)
    pot_dup_corrs  = np.array(np.array(pot_dup_meta)[:, 4], dtype=float)
    pot_dup_dists  = np.array(np.array(pot_dup_meta)[:, 5], dtype=float)

    n_pot_dups   = pot_dup_inds.shape[0]

    dup_mdata  = []
    dup_mdata += [['ind 1', 'ind 2']+[ki+' %d'%ii_kk for kk in keys_to_print for ki in kk.split('|')  for ii_kk in [1, 2]]]

    # loop through the potential duplicates.
    for i_pot_dups, (ii, jj) in enumerate(pot_dup_inds):
        dup_mdata_row    = [ii,jj]
        # data and metadata associated with the two potential duplicates
        id_1, id_2       = df['datasetId'].iloc[[ii, jj]]
        time_1, time_2   = df['year'].iloc[[ii, jj]]
        time_12, int_1, int_2 = np.intersect1d(time_1, time_2, return_indices=True) # possibly need to round these time series?
        data_1           = np.array(df['paleoData_values'].iloc[ii])
        data_2           = np.array(df['paleoData_values'].iloc[jj])

        lat_1, lat_2     = df['geo_meanLat'].iloc[[ii, jj]]
        lon_1, lon_2     = df['geo_meanLon'].iloc[[ii, jj]]
        site_1, site_2   = df['geo_siteName'].iloc[[ii, jj]]
        or_db_1, or_db_2 = df['originalDatabase'].iloc[[ii, jj]]

        print('> %d/%d'%(i_pot_dups, n_pot_dups), #set_txt, 
               id_1, id_2, #var_1,var_2, season_1,season_2,
               pot_dup_dists[i_pot_dups], pot_dup_corrs[i_pot_dups],
               sep=',')

        # Plot observations
        #-----------------------------------------------------

        # Metadata display parameters
        fig, dup_mdata_row = dup_plot(df, ii, jj, id_1, id_2, time_1, time_2, time_12, data_1, 
                                      data_2, int_1, int_2, pot_dup_corrs[i_pot_dups],
                                      keys_to_print, dup_mdata_row)


        dup_mdata += [dup_mdata_row]
        plt.show(block=False)
        if save_figures:
            save_fig(fig, f'%03d_{id_1}_{id_2}'%i_pot_dups+'_'+f'_{ii}_{jj}', dir='%s/dup_detection'%df.name, figformat='pdf')
        if not display:
            plt.close(fig)


        print('=== POTENTIAL DUPLICATE %d/%d'%(i_pot_dups, n_pot_dups)+': %s+%s ==='%(df['datasetId'].iloc[ii], df['datasetId'].iloc[jj]))
    return