使用Tensorflow.keras创建简单模型

1. 使用keras的API配置训练

训练代码：

import tensorflow as tf
from tensorflow import keras
import numpy as np
from matplotlib import pyplot as plt

np.random.seed(42)  # 设置numpy随机数种子
tf.random.set_seed(42) # 设置tensorflow随机数种子

# 生成训练数据
x = np.linspace(-1, 1, 100)
y = x * x + 1 + np.random.rand(100)*0.1  # y=x^2+1 + 随机噪声
x_train = np.expand_dims(x, 1)  # 将一维数据扩展为二维
y_train = np.expand_dims(y, 1)  # 将一维数据扩展为二维
plt.plot(x, y, '.')  # 画出训练数据

# 创建模型：输入层1个节点，隐藏层10个节点，输出层1个节点
model = keras.Sequential()
model.add(keras.Input(shape=(1,)))
model.add(keras.layers.Dense(10, keras.activations.relu))
model.add(keras.layers.Dense(1))

# 打印模型架构
print(model.summary())

# 模型配置（配置优化器，损失函数等）
model.compile(optimizer='sgd',  # 优化器
              loss='mse')  # 损失函数

# 模型训练
model.fit(x_train,
          y_train,
          epochs=1000)  # 迭代次数

# 预测
y_pre = model.predict(x_train)

# 画出预测值
plt.plot(x, y_pre)
plt.show() 

打印结果：

Model: "sequential"
_________________________________________________________________ 
Layer (type)                 Output Shape              Param #   
================================================================= 
dense (Dense)                (None, 10)                20        
_________________________________________________________________
dense_1 (Dense)              (None, 1)                 11        
================================================================= 
Total params: 31
Trainable params: 31
Non-trainable params: 0
_________________________________________________________________ 

2. 手动进行前向和反向传播

训练代码：

import tensorflow as tf
from tensorflow import keras
import numpy as np
from matplotlib import pyplot as plt

np.random.seed(42)  # 设置numpy随机数种子
tf.random.set_seed(42)  # 设置tensorflow随机数种子

# 生成训练数据
x = np.linspace(-1, 1, 100)
x = x.astype('float32')
y = x * x + 1 + np.random.rand(100)*0.1  # y=x^2+1 + 随机噪声
x_train = np.expand_dims(x, 1)  # 将一维数据扩展为二维
y_train = np.expand_dims(y, 1)  # 将一维数据扩展为二维

plt.plot(x, y, '.')  # 画出训练数据

def create_model():
    inputs = keras.Input((1,))
    x = keras.layers.Dense(10, activation='relu')(inputs)
    outputs = keras.layers.Dense(1)(x)
    model = keras.Model(inputs=inputs, outputs=outputs)
    return model


model = create_model()  # 创建一个模型

loss_fn = keras.losses.MeanSquaredError()  # 定义损失函数
optimizer = keras.optimizers.SGD()  # 定义优化器

epochs = 1000  # 训练次数
for epoch in range(epochs):
    with tf.GradientTape() as tape:
        y_pred = model(x_train, training=True)  # 前向传播，注意不要忘了training=True
        loss = loss_fn(y_train, y_pred)  # 计算损失
    grads = tape.gradient(loss, model.trainable_variables)  # 计算梯度
    optimizer.apply_gradients(zip(grads, model.trainable_variables))  # 使用优化器进行反向传播

    print('epoch:', epoch+1, '\t', 'loss:', loss.numpy())  # 打印训练信息

# 预测
y_pre = model.predict(x_train)

# 画出预测值
plt.plot(x, y_pre.squeeze())
plt.show()