The Duplicate generator is a libray to create synthetic personal demographic data such as given name, surname, date of birth and sex and introduce duplicates of those records with errors. Those data can be used to test or evaluate record linkage and deduplication algorithm.

This project is a fork and an udpated version of the Freely Extensible Biomedical Record Linkage (FEBRL) dataset generator developed by Agus Pudjijono and Peter Christen in December 2008. For more detailed description please consult the website of the AUSTRALIAN NATIONAL UNIVERSITY(ANU) Department of Computer Science or see the paper: Accurate Synthetic Generation of Realistic Personal Information.

In this version we upgraded the initial code to python 3.6 , added pandas, argparse and numpy, decoupled the configuration from the code and re-designed the library api to make it easy and simple to generate a customizable synthetic duplicate personal dataset.

import duplicategenerator

dupgen =  duplicategenerator.DuplicateGen(
            num_org_records = 4000,
            num_dup_records = 1000,
            max_num_dups = 1,
            max_num_field_modifi= 1,
            max_num_record_modifi= 1,
            prob_distribution = "uniform",
            type_modification= "all",
            verbose_output = False,
            culture = "eng",
            attr_file_name = './attr_config_file.example.json',
            field_names_prob = {'culture' : 0,
                                'sex': 0,
                                'given_name':0.3,
                                'surname':0.3, 
                                'date_of_birth':0.15,
                                'phone_number':0.2,
                                'national_identifier':0.05}
        )


df = dupgen.generate("dataframe")
df

   

This will generate a panda dataframe with 5000 records (4000 originals and 1000 duplicates) . The dataset generated is fully customizable using the parameters :

• num_org_records Number of original records to be created
• num_dup_records Number of duplicate records to be created
• max_num_dups The maximum number of duplicates that can be created for one original record
• max_num_field_modifi The maximum number of modifications per field
• max_num_record_modifi The maximum number of modifications per record
• prob_distribution The probability distribution used to create the duplicates (i.e the number of duplicates for one original)
• type_modification Select the modification/error types that will be used when duplicates are created
• verbose_output Show logging information
• culture Select country or language for names
• attr_file_name Configuration file with errors probability for each field
• field_names_prob List of fields in the dataset with the probability to select for modifications/errors when creating duplicates

python -m duplicategenerator ./test/test5.csv 4000 1000 5 2 2 uniform all --culture eng --config_file ./duplicategenerator/
config/attr_config_file.example.json

To use the command line you must provide :

• output_file CSV file name to save the data

• Real-world Data is Dirty: Data Cleansing and The Merge/Purge Problem (1998)
• Accurate Synthetic Generation of Realistic Personal Information.

NOTE: This is a beta product. Most of the orginal authors functionalities to model phonetic, OCR and typographical modifications are in an 'alpha' state and have received limited testing.