This is a python implementation of SigDirect [1] classifier. The classifier does not have any specific hyper-parameter to tune and the only one is the p-value of the statistical significance, which is set to 0.05 (you can change it in the config file, though this is the value used in most scientific works)
You first need to use the requirements.txt file to install the dependencies:
pip3 install -r requirements.txt
You can test the dataset using one of the provided UCI datasets:
python3 sigdirect_test.py iris
In order to use the classifier, you can call the code similar to scikit-learn classifiers:
You should instantiate it first, and then call the fit
method to train a model.
To get predictions, you can either call predict
or predict_proba
methods where
the former will provide classes and the latter will provide the probability distributions.
The input to fit method should be a 2-d numpy array where each feature can be either 0 or 1. For labels, a 1-d array should be used where each element should also be integers (starting from 0 to n-1 for n classes.)
Note: You can provide get_logs=std.out
or get_logs=std.err
as an argument to the constructor
of the classifier so it will print some logs about creating the model.
[1]: Li, Jundong, and Osmar R. Zaiane. "Exploiting statistically significant dependent rules for associative classification." Intelligent Data Analysis 21.5 (2017): 1155-1172.