rl-acc is a supervised method to learn and assess the accuracy of record linkage. It requires a labeled dataset in order to work. The dataset must have at least three columns: the label (y) and two coluns with information from the two liked datasets.
The first step before running rl-acc is to edit the config file acording to your data. Assuming you have three columns (y, name_a, and name_b [assuming a and b were the two liked datasets]) you would have the following config file:
[DEFAULT]
DBA = a
DBB = b
IndexScore =
Y = y
[0]
Id = name
Type = name
IndexA = name_a
IndexB = name_b
Features = jaro
In the config above,
Y means the name of the column in the data containing the label,
IndexA and IndexB the name of the column containing the variable to be used to extract the features.
Features defines the features extracted for this column, which can also be a list, separated by comma (ie, jaro,hamming).
Each entry of Features must be defined in featuresExtractor.py inside extractFeatures function.
Once the config file is configured properly,
you can use the main.ipynb notebook to run rl-acc.
The notebook is self expanatory.
If you have any questions, please contact me at gcgbarbosa@gmail.com