fork download 
  1. # your code goes here
  2. import numpy as np
  3.  
  4. class TrainableXNOR:
  5.     def __init__(self, learning_rate=0.1):
  6.         # Инициализация весов случайными значениями
  7.         np.random.seed(42)
  8.         self.W1 = np.random.randn(2, 2) * 0.01
  9.         self.b1 = np.zeros((1, 2))
  10.         self.W2 = np.random.randn(2, 1) * 0.01
  11.         self.b2 = np.zeros((1, 1))
  12.         self.lr = learning_rate
  13.  
  14.     def sigmoid(self, x):
  15.         return 1 / (1 + np.exp(-np.clip(x, -700, 700)))
  16.  
  17.     def sigmoid_derivative(self, x):
  18.         return x * (1 - x)
  19.  
  20.     def forward(self, X):
  21.         self.z1 = np.dot(X, self.W1) + self.b1
  22.         self.a1 = self.sigmoid(self.z1)
  23.         self.z2 = np.dot(self.a1, self.W2) + self.b2
  24.         self.a2 = self.sigmoid(self.z2)
  25.         return self.a2
  26.  
  27.     def backward(self, X, y, output):
  28.         m = X.shape[0]
  29.  
  30.         # Градиент выходного слоя
  31.         dz2 = output - y
  32.         dW2 = (1/m) * np.dot(self.a1.T, dz2)
  33.         db2 = (1/m) * np.sum(dz2, axis=0, keepdims=True)
  34.  
  35.         # Градиент скрытого слоя
  36.         da1 = np.dot(dz2, self.W2.T)
  37.         dz1 = da1 * self.sigmoid_derivative(self.a1)
  38.         dW1 = (1/m) * np.dot(X.T, dz1)
  39.         db1 = (1/m) * np.sum(dz1, axis=0, keepdims=True)
  40.  
  41.         # Обновление параметров
  42.         self.W2 -= self.lr * dW2
  43.         self.b2 -= self.lr * db2
  44.         self.W1 -= self.lr * dW1
  45.         self.b1 -= self.lr * db1
  46.  
  47.     def train(self, X, y, epochs=10000):
  48.         losses = []
  49.         for epoch in range(epochs):
  50.             output = self.forward(X)
  51.             loss = np.mean((output - y) ** 2)
  52.             losses.append(loss)
  53.             self.backward(X, y, output)
  54.  
  55.             if epoch % 1000 == 0:
  56.                 print(f"Epoch {epoch}, Loss: {loss:.6f}")
  57.         return losses
  58.  
  59. # Данные для обучения
  60. X = np.array([[0, 0], [0, 1], [1, 0], [1, 1]])
  61. y = np.array([[1], [0], [0], [1]])  # XNOR
  62.  
  63. # Обучение и тестирование
  64. network = TrainableXNOR(learning_rate=0.5)
  65. losses = network.train(X, y, epochs=10000)
  66.  
  67. print("\nРезультаты после обучения:")
  68. print("Input  Target  Output    Predicted")
  69. for i in range(len(X)):
  70.     output = network.forward(X[i].reshape(1, -1))
  71.     predicted = 1 if output > 0.5 else 0
  72.     print(f"{X[i]}   {y[i][0]}      {output[0][0]:.4f}    {predicted}")
Success #stdin #stdout 2.11s 42132KB
comments (?)
stdin 
Standard input is empty
stdout 
Epoch 0, Loss: 0.250000
Epoch 1000, Loss: 0.250000
Epoch 2000, Loss: 0.250000
Epoch 3000, Loss: 0.250000
Epoch 4000, Loss: 0.250000
Epoch 5000, Loss: 0.250000
Epoch 6000, Loss: 0.250000
Epoch 7000, Loss: 0.250000
Epoch 8000, Loss: 0.250000
Epoch 9000, Loss: 0.250000

Результаты после обучения:
Input  Target  Output    Predicted
[0 0]   1      0.5000    1
[0 1]   0      0.5000    0
[1 0]   0      0.5000    1
[1 1]   1      0.5000    0
https://ideone.com/IUSQyG
language:
Python 3 (python  3.12)
created:
1 day ago
visibility:
public

Share or Embed source code

Discover > Sphere Engine API

The brand new service which powers Ideone!

Discover > IDE Widget

Widget for compiling and running the source code in a web browser!