main.py

"""
@author: Farzam.Taghipour
https://github.com/FarzamTP/Logistic-Regression
"""
from __future__ import print_function

from LogisticRegression import LogisticRegression
from sklearn.linear_model import LogisticRegression as skLR

import numpy as np


# reads file and generates X, y.
def read_data_from_file(file_path):
    with open(file_path, 'r') as f:
        X = []
        y = []
        for line in f.readlines():
            X.append([float(line.split(',')[0]), float(line.split(',')[1])])
            y.append(float(line.split(',')[2]))
    return X, y


# calling read from file function that returns X, y as numpy arrays.
X, y = read_data_from_file('data.txt')

# converting X , y to numpy arrays
X = np.asarray(X)
y = np.asarray(y)

# prints X, y shapes.
print('X shape:', X.shape)
print('y shape:', y.shape)

# declaring a list in size of selected dataset containing 0 to size of dataset.
indices = list(range(len(X)))

# shuffling indices array.
np.random.shuffle(indices)

# number of data samples we want to train
num_training_samples = 80

# splitting train data arrays according to indices array.
x_train = X[indices[:num_training_samples]]
y_train = y[indices[:num_training_samples]]

# splitting test data arrays according to indices array.
x_test = X[indices[num_training_samples:]]
y_test = y[indices[num_training_samples:]]

# Defining sklearn.linear model with lbfgs solver
sk_model = skLR(C=1000000, solver='lbfgs')

# fitting sklearn.linear.LogisticRegression model over train set.
sk_model.fit(x_train, y_train)

# declaring my object of class logistic regression.
# LogisticRegression(X, y, num_iter=100, learning_rate=0.01, verbose=True)
model = LogisticRegression(num_iter=1000000, lr=0.01, Lambda=0, verbose=True)

# printing number of train set data points.
print()
print("|================================================|")
print("|============== Logistic Regression =============|")
print("|================================================|")
print()
print('Training %s numbers of train data points...' % num_training_samples)
print()

# plotting data.
model.plot_data()

# fitting model over train set.
model.fit(x_train, y_train)

# printing Theta values of trained sklearn.linear model.
print()
print('Sklearn coefficient:\n', sk_model.coef_)
print(sk_model.intercept_)

# printing Theta values of trained LogisticRegression model.
print('My coefficient:\n', model.theta)
print()

# plotting cost values over iterations.
model.plot_cost()

# plot line.
model.plot_line(X, y)

# predicting sklearn.linear model over test set.
sk_predict = sk_model.predict(x_test)

# predicting LogisticRegression model over test set.
my_predict = model.predict(x_test, threshold=0.5)

# printing original labels of test set.
# print('True Labels:\n', y_test)

# printing sklearn.linear model prediction.
# print('Sklearn.linear model predicts:\n', sk_predict)

# printing LogisticRegression model prediction.
# print('LogisticRegression model predict:\n', my_predict)

# comparing prediction of sklearn.linear model and original test set labels.
# calculating accuracy:
accuracy = sum(sk_predict == y_test) / (X.shape[0] - num_training_samples)
print("Sklearn.linear model Accuracy: %.3f" % accuracy + " %")
print()

# comparing prediction of LogisticRegression model and original test set labels.
# calculating accuracy:
my_accuracy = sum(my_predict == y_test) / (X.shape[0] - num_training_samples)
print("LogisticRegression Accuracy: %.3f" % my_accuracy + " %")