Keras练习：线性回归

Keras是高层神经网络API，后端可基于Tensorflow运行。

这里创建一个简单数据集，做线性回归，感受一下Keras的便捷。

Regression

import numpy as np
import matplotlib.pyplot as plt
from keras.models import Sequential
from keras.layers import Dense


np.random.seed(1327)


def create_data():
    x = np.linspace(-1, 1, 200)
    np.random.shuffle(x)
    y = 0.5 * x + 2 + np.random.normal(0, 0.05, (200,))
    return x, y


def build_model():
    model = Sequential()
    model.add(Dense(input_dim=1, units=1))
    model.compile(loss='mse', optimizer='sgd')
    return model


def train(model, x, y):
    print('Training........')
    for step in range(1001):
        cost = model.train_on_batch(x, y)
        if step % 100 == 0:
            print('COST:', cost)


def test(model, x, y):
    print('\nTesting.......')
    cost = model.evaluate(x_test, y_test, batch_size=40)
    print('TEST COST: ', cost)
    w, b = model.layers[0].get_weights()
    print('Weights=', w, 'biases=', b)


if __name__ == '__main__':
    x, y = create_data()
    # plt.scatter(x, y)
    # plt.show()

    x_train = x[:160]
    y_train = y[:160]
    x_test = x[160:]
    y_test = y[160:]

    model = build_model()
    train(model, x_train, y_train)
    test(model, x_test, y_test)

    y_predict = model.predict(x_test)
    plt.scatter(x_test, y_test)
    plt.plot(x_test, y_predict)
    plt.show()

输出：

Training........
COST: 4.0496254
COST: 0.08321373
COST: 0.0063530593
COST: 0.0031990125
COST: 0.0026882463
COST: 0.002564282
COST: 0.0025330638
COST: 0.0025251824
COST: 0.002523192
COST: 0.0025226888
COST: 0.0025225622

Testing.......
40/40 [==============================] - 0s 1ms/step
TEST COST:  0.0018893185770139098
Weights= [[0.5145617]] biases= [1.9962281]