crossEntropy.py

import matplotlib.pyplot as plt
import numpy as np

def generate_data_or():
    x = [[0,0], [1,1], [1,0], [0,1]]
    y = [0, 1, 1, 1]
    return x, y

# y zu oneHotVector umwandeln

def to_one_hot(y, num_classes):
    y_one_hot = np.zeros(shape=(len(y), num_classes)) # 4 x 2
    for i, y_i in enumerate(y):
        #y_oh = [1 if c == y_i else 0 for c in range(num_classes)]
        y_oh = np.zeros(shape=num_classes)
        y_oh[y_i] = 1
        y_one_hot[i] = y_oh
    return y_one_hot

x, y = generate_data_or()
y = to_one_hot(y, num_classes = 2)
print(y)

p1 = [0.223, 0.613]
p2 = [0.145, -0.75]
p3 = [0.45, 0.2]
p4 = [0.66, 0.19]
y_pred = np.array([p1, p2, p3, p4])

def softmax(y_pred):
    y_softmax = np.zeros(shape=y_pred.shape)
    for i in range(len(y_pred)):
        exps = np.exp(y_pred[i])
        y_softmax[i] = exps / np.sum(exps)
    return y_softmax

print(y_pred)
y_pred = softmax(y_pred)
print(y_pred)


def cross_entropy(y_true, y_pred):
    num_samples = y_pred.shape[0]
    num_classes = y_pred.shape[1]
    loss = 0.0
    for y_t, y_p in zip(y_true, y_pred):
        for c in range(num_classes):
            loss -= y_t[c] * np.log(y_p[c])
    return loss / num_samples

loss = cross_entropy(y, y_pred)
print(loss)