《PyTorch深度学习实践》梯度下降法

课程请见 《PyTorch深度学习实践》
用numpy适配了一下，均改成了numpy.ndarray类型

# Numpy
import numpy
# For plotting
import matplotlib.pyplot as plt
from matplotlib.pyplot import figure


def forward(w, x):
    return x * w


def cost(x_cor, y_cor, w):
    y_hat = forward(w, x_cor)
    loss = (y_hat - y_cor) ** 2
    return loss.sum() / len(x_cor)


def gradient(x_cor, y_cor, w):
    grad = 2 * x_cor * (w * x_cor - y_cor)
    return grad.sum() / len(x_cor)


x_data = numpy.array([1.0, 2.0, 3.0])
y_data = numpy.array([2.0, 4.0, 6.0])
num_epochs = 100
lr = 0.01
w_train = numpy.array([1.0])
epoch_cor = []
loss_cor = []
for epoch in range(num_epochs):
    mse_loss = cost(x_data, y_data, w_train)
    loss_cor.append(mse_loss)
    w_train -= lr * gradient(x_data, y_data, w_train)
    epoch_cor.append(epoch + 1)

plt.figure()
plt.plot(epoch_cor, loss_cor, c='b')
plt.xlabel('Epoch')
plt.ylabel('Cost')
plt.show()

最终效果和老师的 一模一样