……

 

1 保存序列模型或函数式模型

In [1]:
# 构建一个简单的模型并训练
from __future__ import absolute_import, division, print_function
import tensorflow as tf
tf.keras.backend.clear_session()
from tensorflow import keras
from tensorflow.keras import layers

inputs = keras.Input(shape=(784,), name='digits')
x = layers.Dense(64, activation='relu', name='dense_1')(inputs)
x = layers.Dense(64, activation='relu', name='dense_2')(x)
outputs = layers.Dense(10, activation='softmax', name='predictions')(x)

model = keras.Model(inputs=inputs, outputs=outputs, name='3_layer_mlp')
model.summary()
(x_train, y_train), (x_test, y_test) = keras.datasets.mnist.load_data()
x_train = x_train.reshape(60000, 784).astype('float32') / 255
x_test = x_test.reshape(10000, 784).astype('float32') / 255

model.compile(loss='sparse_categorical_crossentropy',
              optimizer=keras.optimizers.RMSprop())
history = model.fit(x_train, y_train,
                    batch_size=64,
                    epochs=1)

predictions = model.predict(x_test)
 
/home/doit/anaconda3/lib/python3.6/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.
  from ._conv import register_converters as _register_converters
 
Model: "3_layer_mlp"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
digits (InputLayer)          [(None, 784)]             0         
_________________________________________________________________
dense_1 (Dense)              (None, 64)                50240     
_________________________________________________________________
dense_2 (Dense)              (None, 64)                4160      
_________________________________________________________________
predictions (Dense)          (None, 10)                650       
=================================================================
Total params: 55,050
Trainable params: 55,050
Non-trainable params: 0
_________________________________________________________________
Train on 60000 samples
60000/60000 [==============================] - 2s 28us/sample - loss: 0.3133
 

1.1 保存整个模型

可以对整个模型进行保存,其保持的内容包括:

  • 该模型的架构
  • 模型的权重(在训练期间学到的)
  • 模型的训练配置(传递给编译的)
  • 优化器及其状态(这使您可以从中断的地方重新启动训练)
In [2]:
import numpy as np
# 模型保存
model.save('the_save_model.h5')
# 导入模型
new_model = keras.models.load_model('the_save_model.h5')
new_prediction = new_model.predict(x_test)
np.testing.assert_allclose(predictions, new_prediction, atol=1e-6) # 预测结果一样
 

1.2 导出为SavedModel文件

SavedModel是Tensorflow对象的独立序列化格式,支持使用Tensorflow Serving server来部署模型,支持其他语言读取。

In [4]:
# 导出为tf的SavedModel文件
model.save('save_model', save_format='tf')
# 从SavedModel文件中导入模型
new_model = keras.models.load_model('save_model')

new_prediction = new_model.predict(x_test)
np.testing.assert_allclose(predictions, new_prediction, atol=1e-6) # 预测结果一样
 
W1013 16:41:11.934549 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer.decay
W1013 16:41:11.935179 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer.learning_rate
W1013 16:41:11.935569 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer.momentum
W1013 16:41:11.935910 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer.rho
W1013 16:41:11.936488 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer's state 'rms' for (root).layer_with_weights-0.kernel
W1013 16:41:11.937078 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer's state 'rms' for (root).layer_with_weights-0.bias
W1013 16:41:11.937702 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer's state 'rms' for (root).layer_with_weights-1.kernel
W1013 16:41:11.938094 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer's state 'rms' for (root).layer_with_weights-1.bias
W1013 16:41:11.938518 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer's state 'rms' for (root).layer_with_weights-2.kernel
W1013 16:41:11.938826 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer's state 'rms' for (root).layer_with_weights-2.bias
W1013 16:41:11.939161 140085562951488 util.py:152] A checkpoint was restored (e.g. tf.train.Checkpoint.restore or tf.keras.Model.load_weights) but not all checkpointed values were used. See above for specific issues. Use expect_partial() on the load status object, e.g. tf.train.Checkpoint.restore(...).expect_partial(), to silence these warnings, or use assert_consumed() to make the check explicit. See https://www.tensorflow.org/alpha/guide/checkpoints#loading_mechanics for details.
W1013 16:41:11.940948 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer
W1013 16:41:11.941336 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer.iter
W1013 16:41:11.941909 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer.decay
W1013 16:41:11.944620 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer.learning_rate
W1013 16:41:11.945564 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer.momentum
W1013 16:41:11.946080 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer.rho
W1013 16:41:11.946562 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer's state 'rms' for (root).layer_with_weights-0.kernel
W1013 16:41:11.946892 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer's state 'rms' for (root).layer_with_weights-0.bias
W1013 16:41:11.947290 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer's state 'rms' for (root).layer_with_weights-1.kernel
W1013 16:41:11.947617 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer's state 'rms' for (root).layer_with_weights-1.bias
W1013 16:41:11.947977 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer's state 'rms' for (root).layer_with_weights-2.kernel
W1013 16:41:11.948273 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer's state 'rms' for (root).layer_with_weights-2.bias
W1013 16:41:11.948586 140085562951488 util.py:152] A checkpoint was restored (e.g. tf.train.Checkpoint.restore or tf.keras.Model.load_weights) but not all checkpointed values were used. See above for specific issues. Use expect_partial() on the load status object, e.g. tf.train.Checkpoint.restore(...).expect_partial(), to silence these warnings, or use assert_consumed() to make the check explicit. See https://www.tensorflow.org/alpha/guide/checkpoints#loading_mechanics for details.
 

SaveModel创建的文件包含:

  • 权重
  • 网络图
 

1.3 仅保存网络结构

仅保持网络结构,这样导出的模型并未包含训练好的参数

In [5]:
# 获取网络结构配置
config = model.get_config()
reinitialized_model = keras.Model.from_config(config)

new_prediction = reinitialized_model.predict(x_test)
assert abs(np.sum(predictions-new_prediction)) >0
 

也可以使用json保存网络结构

In [6]:
# 将网络结构导出为json格式
json_config = model.to_json()
reinitialized_model = keras.models.model_from_json(json_config)

new_prediction = reinitialized_model.predict(x_test)
assert abs(np.sum(predictions-new_prediction)) >0
 

1.4 仅保存网络权重参数

In [7]:
# 获取网络权重
weights = model.get_weights()
# 对网络权重进行赋值
model.set_weights(weights)
 

可以把结构和参数保存结合起来

In [8]:
config = model.get_config()
weights = model.get_weights()

new_model = keras.Model.from_config(config) # config只能用keras.Model的这个api
new_model.set_weights(weights)

new_predictions = new_model.predict(x_test)
np.testing.assert_allclose(predictions, new_predictions, atol=1e-6)
 

1.5 完整的模型保存方法

In [10]:
# 导出网络结构和权重
json_config = model.to_json()
with open('model_config.json', 'w') as json_file:
    json_file.write(json_config)
model.save_weights('path_to_my_weights.h5')
# 载入网络结构和权重
with open('model_config.json') as json_file:
    json_config = json_file.read()
new_model = keras.models.model_from_json(json_config)
new_model.load_weights('path_to_my_weights.h5')

new_predictions = new_model.predict(x_test)
np.testing.assert_allclose(predictions, new_predictions, atol=1e-6)
In [11]:
# 当然也可以一步到位
model.save('path_to_my_model.h5')
del model
model = keras.models.load_model('path_to_my_model.h5')
 
W1013 16:49:21.690537 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer
W1013 16:49:21.691003 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer.iter
W1013 16:49:21.691583 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer.decay
W1013 16:49:21.691886 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer.learning_rate
W1013 16:49:21.692354 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer.momentum
W1013 16:49:21.692843 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer.rho
W1013 16:49:21.693265 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer's state 'rms' for (root).layer_with_weights-0.kernel
W1013 16:49:21.693663 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer's state 'rms' for (root).layer_with_weights-0.bias
W1013 16:49:21.693997 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer's state 'rms' for (root).layer_with_weights-1.kernel
W1013 16:49:21.694370 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer's state 'rms' for (root).layer_with_weights-1.bias
W1013 16:49:21.694779 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer's state 'rms' for (root).layer_with_weights-2.kernel
W1013 16:49:21.695171 140085562951488 util.py:144] Unresolved object in checkpoint: (root).optimizer's state 'rms' for (root).layer_with_weights-2.bias
W1013 16:49:21.695466 140085562951488 util.py:152] A checkpoint was restored (e.g. tf.train.Checkpoint.restore or tf.keras.Model.load_weights) but not all checkpointed values were used. See above for specific issues. Use expect_partial() on the load status object, e.g. tf.train.Checkpoint.restore(...).expect_partial(), to silence these warnings, or use assert_consumed() to make the check explicit. See https://www.tensorflow.org/alpha/guide/checkpoints#loading_mechanics for details.
 

1.6 权重保存格式

有.h5或.keras后缀时保存为keras HDF5格式文件,否则默认为TensorFlow Checkpoint格式文件。可以使用save_format显式确定。

In [12]:
model.save_weights('weight_tf_savedmodel')
model.save_weights('weight_tf_savedmodel.h5')
In [13]:
model.save_weights('weight_tf_savedmodel_tf', save_format='tf')
model.save_weights('weight_tf_savedmodel_h5', save_format='h5')
 

1.7 子类模型权重保存

子类模型的结构无法保存和序列化,只能保持参数

In [16]:
# 构建模型
class ThreeLayerMLP(keras.Model):
  
    def __init__(self, name=None):
        super(ThreeLayerMLP, self).__init__(name=name)
        self.dense_1 = layers.Dense(64, activation='relu', name='dense_1')
        self.dense_2 = layers.Dense(64, activation='relu', name='dense_2')
        self.pred_layer = layers.Dense(10, activation='softmax', name='predictions')

    def call(self, inputs):
        x = self.dense_1(inputs)
        x = self.dense_2(x)
        return self.pred_layer(x)

def get_model():
    return ThreeLayerMLP(name='3_layer_mlp')

model = get_model()
 

首先,无法保存从未使用过的子类模型。

这是因为需要在某些数据上调用子类模型才能创建其权重。

In [17]:
# 训练模型
(x_train, y_train), (x_test, y_test) = keras.datasets.mnist.load_data()
x_train = x_train.reshape(60000, 784).astype('float32') / 255
x_test = x_test.reshape(10000, 784).astype('float32') / 255

model.compile(loss='sparse_categorical_crossentropy',
              optimizer=keras.optimizers.RMSprop())
history = model.fit(x_train, y_train,
                    batch_size=64,
                    epochs=1)
 
Train on 60000 samples
60000/60000 [==============================] - 2s 26us/sample - loss: 0.3128
 

推荐的保存子类模型的方法是使用save_weights创建TensorFlow SavedModel检查点。

该检查点将包含与模型关联的所有变量的值:

  • 图层的权重
  • 优化器的状态
  • 与有状态模型指标关联的任何变量
In [18]:
# 保持权重参数
model.save_weights('my_model_weights', save_format='tf')

# 输出结果,供后面对比

predictions = model.predict(x_test)
first_batch_loss = model.train_on_batch(x_train[:64], y_train[:64])
 

要还原模型,将需要访问创建模型对象的代码。

请注意,为了恢复优化器状态和任何有状态度量的状态,应该先编译模型(使用与以前完全相同的参数)。

In [19]:
# 读取保存的模型参数
new_model = get_model()
new_model.compile(loss='sparse_categorical_crossentropy',
                  optimizer=keras.optimizers.RMSprop())

#new_model.train_on_batch(x_train[:1], y_train[:1])

new_model.load_weights('my_model_weights')

new_predictions = new_model.predict(x_test)
np.testing.assert_allclose(predictions, new_predictions, atol=1e-6)


new_first_batch_loss = new_model.train_on_batch(x_train[:64], y_train[:64])
assert first_batch_loss == new_first_batch_loss
In [ ]:
 
 posted on 2020-10-20 10:43  大码王  阅读(726)  评论(0编辑  收藏  举报
复制代码