usage_example.py

import numpy as np
import sys
sys.path.append('../efc')
import pandas as pd
from classification_functions import *
from generic_discretize import *
from sklearn.metrics import confusion_matrix
from sklearn.datasets import load_breast_cancer, load_wine
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier


'''Usage example of Single-class EFC'''
data = load_breast_cancer(as_frame=True) # load toy dataset from scikit-learn (binary targets)
X_train, X_test, y_train, y_test = train_test_split(data.data, data.target, test_size=0.33, shuffle=False) # split dataset into training and testing sets
intervals = get_intervals(X_train, 10) # get discretization intervals from train set
X_train = discretize(X_train, intervals) # discretize train
X_test = discretize(X_test, intervals) # discretize test

idx_abnormal = np.where(y_train == 1)[0] # find abnormal samples indexes in the training set
X_train.drop(idx_abnormal, axis=0, inplace=True) # remove abnormal samples from training (EFC trains with only benign instances)
y_train.drop(idx_abnormal, axis=0, inplace=True) # remove the corresponding abonrmal training targets

#EFC's hyperparameters
Q = X_test.values.max()
LAMBDA = 0.5 # pseudocount parameter

coupling, h_i, cutoff = OneClassFit(np.array(X_train), Q, LAMBDA) # train model
y_predicted, energies = OneClassPredict(np.array(X_test), coupling, h_i, cutoff, Q) # test model

# colect results
print('Single-class results')
print(confusion_matrix(y_test, y_predicted))


'''Usage example of Multi-class EFC'''
data = load_wine(as_frame=True) # load toy dataset from scikit-learn (binary targets)
X_train, X_test, y_train, y_test = train_test_split(data.data, data.target, test_size=0.33, shuffle=False) # split dataset into training and testing sets
intervals = get_intervals(X_train, 10) # get discretization intervals from train set
X_train = discretize(X_train, intervals) # discretize train
X_test = discretize(X_test, intervals) # discretize test

#EFC's hyperparameters
Q = X_test.values.max()
LAMBDA = 0.5 # pseudocount parameter

coupling_matrices, h_i_matrices, cutoffs_list = MultiClassFit(np.array(X_train), np.array(y_train), Q, LAMBDA) # train model
y_predicted = MultiClassPredict(np.array(X_test), coupling_matrices, h_i_matrices, cutoffs_list, Q, np.unique(y_train)) # test model

# colect results
print('Multi-class results')
print(confusion_matrix(y_test, y_predicted))