在之前,我们已经学习了Keras高层接口的应用,主要是compile()和fit()函数,前者装配模型,后者控制训练流程,十分的方便;
networks.compile(optimizers = optimizers.Adam(lr=0.01),#定义优化器
loss = tf.losses.CategoricalCrossentropy(from_logits=True), #指定Loss
metrics = ['accuracy'] #指定测试指标
)
network.fit(db,epoch=10) #指定迭代次数
network.fit(db,epoch=10,validation_data=ds_val,validation_freq=2) #指定测试
netword.evaluate(tt_val) #进行验证
其次,我们还学习了使用Sequential() 来建立神经网络的容器,以及如何自定义网络,包括层Dense和模型Model;
class MyDense(layers.Layer): #自己实现一个线性层Dense,继承自layers.Layer
def __init__(self,inp_dim,outp_dim): #实现初始化方法,必须要有
super(MyDense,self).__init__() #调用母类的初始化函数,必须要有
#创建遍历
self.kernel = self.add_variable('w',[inp_dim,outp_dim])
self.bias = self.add_variable('b',[outp_dim])
def call(self, inputs, training=None): #实现call方法,必须要有
out = inputs @ self.kernel + self.bias #矩阵相乘再相加
return out
class MyModel(tf.keras.Model): #自己实现一个模型,必须继承自tf.keras.Model
def __init__(self): #实现初始化方法,必须要有
super(MyModel,self).__init__() #调用母类的初始化函数,必须要有
self.fc1 = MyDense(28*28,256)
self.fc2 = MyDense(256, 128)
self.fc3 = MyDense(128, 64)
self.fc4 = MyDense(64, 32)
self.fc5 = MyDense(32, 10)
def call(self, inputs, training=None):
x = self.fc1(inputs)
x = tf.nn.relu(x) #激活函数,一遍一遍的通过,从而形成新的x
x = self.fc2(inputs)
x = tf.nn.relu(x)
x = self.fc3(inputs)
x = tf.nn.relu(x)
x = self.fc4(inputs)
x = tf.nn.relu(x)
x = self.fc5(inputs) #最后一遍 一般不通过激活函数
return x
还学习了模型的保存与加载,用于在模型训练时保存模型的各种数据,以免意外丢失导致重新训练;
model.save_weight('保存路径') #保存
model = create_model() #加载
model.load_weight('保存路径')
现在我们结合上述所有技术来进行实战,将这些技术运用到真实的数据训练中去,对cifar10进行简单的训练,并查看每次迭代的精确度;
代码如下:
import tensorflow as tf
from tensorflow.keras import datasets, layers, optimizers, Sequential, metrics
from tensorflow import keras
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
def preprocess(x, y): #预处理数据函数
# x取值范围:[0~255] => [-1~1]
x = 2 * tf.cast(x, dtype=tf.float32) / 255. - 1.
y = tf.cast(y, dtype=tf.int32)
return x,y
batchsz = 128
# [50k, 32, 32, 3], [10k, 1]
#(x, y), (x_val, y_val) = datasets.cifar10.load_data()
(x, y), (x_val, y_val) = datasets.mnist.load_data() #加载训练集和测试集
y = tf.squeeze(y) #压缩tensor,将其中所有大小为1的维度删除
y_val = tf.squeeze(y_val)
y = tf.one_hot(y, depth=10) # [50k, 10]
y_val = tf.one_hot(y_val, depth=10) # [10k, 10]
print('datasets:', x.shape, y.shape, x_val.shape, y_val.shape, x.min(), x.max())
train_db = tf.data.Dataset.from_tensor_slices((x,y))
train_db = train_db.map(preprocess).shuffle(10000).batch(batchsz) #预处理,调用preprocess;
# map():函数映射;shuffle():随机打乱;batch():批处理;
test_db = tf.data.Dataset.from_tensor_slices((x_val, y_val))
test_db = test_db.map(preprocess).batch(batchsz)
sample = next(iter(train_db))
print('batch:', sample[0].shape, sample[1].shape)
class MyDense(layers.Layer): #自己实现一个线性层Dense,继承自layers.Layer
# to replace standard layers.Dense()
def __init__(self, inp_dim, outp_dim):
super(MyDense, self).__init__()
self.kernel = self.add_variable('w', [inp_dim, outp_dim]) #创建参数
# self.bias = self.add_variable('b', [outp_dim])
def call(self, inputs, training=None):
x = inputs @ self.kernel
return x
class MyNetwork(keras.Model): #自己实现一个模型,必须继承自tf.keras.Model
def __init__(self):
super(MyNetwork, self).__init__()
self.fc1 = MyDense(32*32*3, 256)
self.fc2 = MyDense(256, 128)
self.fc3 = MyDense(128, 64)
self.fc4 = MyDense(64, 32)
self.fc5 = MyDense(32, 10)
def call(self, inputs, training=None):
#inputs输入: [b, 32, 32, 3]
x = tf.reshape(inputs, [-1, 32*32*3])
# [b, 32*32*3] => [b, 256]
x = self.fc1(x)
x = tf.nn.relu(x) #激活函数,一遍一遍的通过,从而形成新的x
# [b, 256] => [b, 128]
x = self.fc2(x)
x = tf.nn.relu(x)
# [b, 128] => [b, 64]
x = self.fc3(x)
x = tf.nn.relu(x)
# [b, 64] => [b, 32]
x = self.fc4(x)
x = tf.nn.relu(x)
# [b, 32] => [b, 10]
x = self.fc5(x)
return x
network = MyNetwork()
#装配network,设置优化器,损失参数等
network.compile(optimizer=optimizers.Adam(lr=1e-3),
loss=tf.losses.CategoricalCrossentropy(from_logits=True),
metrics=['accuracy'])
#设置训练流程,设置训练集、训练迭代数、测试集
network.fit(train_db, epochs=15, validation_data=test_db, validation_freq=1)
#模型保存,只保存了权值等
network.evaluate(test_db)
network.save_weights('ckpt/weights.ckpt')
del network
print('saved to ckpt/weights.ckpt')
#模型加载
network = MyNetwork()
network.compile(optimizer=optimizers.Adam(lr=1e-3),
loss=tf.losses.CategoricalCrossentropy(from_logits=True),
metrics=['accuracy'])
network.load_weights('ckpt/weights.ckpt')
print('loaded weights from file.')
network.evaluate(test_db)