在Keras中保存模型的正确方法 [英] Proper Way to Save Model in Keras

问题描述

我一直在尝试在研究中实施转移学习。我决定使用 keras.applications 中提供的VGG16。

I have been trying to implement transfer learning in my research. I have decided to go with VGG16 as provided in keras.applications.

我按以下方式加载模型并冻结其权重：

I load the model and freeze its weights as follows:

vgg16 = VGG16(weights='imagenet', include_top=False,input_shape=(img_rows, img_cols, 3), pooling = None)
for layer in vgg16.layers:
    layer.trainable = False

我然后添加顶层进行分类：

I then add top layers for classification:

model = Sequential()
model.add(vgg16)
model.add(Flatten())
model.add(Dense(1024, activation='relu'))
model.add(Dense(1024, activation='relu'))
model.add(Dense(num_classes, activation='softmax'))

我编译并训练数据模型以首先加热顶层：

I compile and train the model on my data to warm up the top layers first:

EPOCHS = 4000

history = model.fit_generator(datagen_train.flow(X_train, y_train, batch_size=100),
                        validation_data = datagen_val.flow(X_val, y_val, batch_size=100),
                        epochs = EPOCHS,
                        steps_per_epoch = np.ceil(len(X_train) / 100),
                        validation_steps = np.ceil(len(X_val) / 100),
                        callbacks=[es, mc]
                       )

我使用常用​​的Keras命令保存模型： save_model 。

I save the model using the usual Keras command: save_model.

我的下一个目标是解冻VGG16的某些顶层并再次训练模型（也称为微调）。但是，在使用 load_model 加载模型后，我发现该模型看起来像未经训练的模型，我在保存测试数据集之前对其进行了测试，并且性能高达70％范围。加载相同的模型后，考虑到我有五个类标签，我发现测试数据集的性能大约为20％，这几乎低于机会。

My next goal is unfreeze some of the top layers of VGG16 and train the model again (a.k.a fine-tune). However, upon loading the model with load_model, I find that the model looks like untrained.I tested it before saving on the test data set and the performance was high in the 70% range. After loading the same model, I find that the performance on the test data set is around 20%, which is almost below chance, considering I have five class labels.

我的 save_model 和 load_model 之间发生了什么命令？

What has happened in between my save_model and load_model commands?

推荐答案

Keras支持一个更简单的界面，可将模型权重和模型体系结构一起保存到单个H5文件中。

Keras supports a simpler interface to save both the model weights and model architecture together into a single H5 file.

使用save.model方法保存模型包括我们需要了解的有关模型的所有信息，包括：

Saving the model with save.model way includes everything we need to know about the model, including:

1. 模型权重。


2. 模型体系结构。


3. 模型编译详细信息（损失和指标）。


4. 模型优化器状态。

1. Model weights.
2. Model architecture.
3. Model compilation details (loss and metrics).
4. Model optimizer state.

然后可以通过调用load_model（）函数并传递文件名来加载保存的模型。函数将返回具有相同架构和权重的模型。

Saved model can then be loaded by calling the load_model() function and passing the filename. The function returns the model with the same architecture and weights.

示例：我运行了一个简单的模型，并使用model.save和用keras的load_model加载了。您可以从此处下载数据集。

Example: I have run a simple model and saved using model.save and the did the load with load_model of keras. You can download the dataset from here.

建立并保存模型：

# MLP for Pima Indians Dataset saved to single file
import numpy as np
from numpy import loadtxt
from keras.models import Sequential
from keras.layers import Dense

# load pima indians dataset
dataset = np.loadtxt("/content/pima-indians-diabetes.csv", delimiter=",")

# split into input (X) and output (Y) variables
X = dataset[:,0:8]
Y = dataset[:,8]

# define model
model = Sequential()
model.add(Dense(12, input_dim=8, activation='relu'))
model.add(Dense(8, activation='relu'))
model.add(Dense(1, activation='sigmoid'))

# compile model
model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])

# Model Summary
model.summary()

# Fit the model
model.fit(X, Y, epochs=150, batch_size=10, verbose=0)

# evaluate the model
scores = model.evaluate(X, Y, verbose=0)
print("%s: %.2f%%" % (model.metrics_names[1], scores[1]*100))

# save model and architecture to single file
model.save("model.h5")
print("Saved model to disk")

输出-

Model: "sequential_3"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense_7 (Dense)              (None, 12)                108       
_________________________________________________________________
dense_8 (Dense)              (None, 8)                 104       
_________________________________________________________________
dense_9 (Dense)              (None, 1)                 9         
=================================================================
Total params: 221
Trainable params: 221
Non-trainable params: 0
_________________________________________________________________
acc: 77.08%
Saved model to disk

加载模型并评估以验证：

# load and evaluate a saved model
from numpy import loadtxt
from keras.models import load_model

# load model
model = load_model('model.h5')

# summarize model.
model.summary()

# load dataset
dataset = loadtxt("pima-indians-diabetes.csv", delimiter=",")

# split into input (X) and output (Y) variables
X = dataset[:,0:8]
Y = dataset[:,8]

# evaluate the model
score = model.evaluate(X, Y, verbose=0)
print("%s: %.2f%%" % (model.metrics_names[1], score[1]*100))

输出-

Model: "sequential_3"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense_7 (Dense)              (None, 12)                108       
_________________________________________________________________
dense_8 (Dense)              (None, 8)                 104       
_________________________________________________________________
dense_9 (Dense)              (None, 1)                 9         
=================================================================
Total params: 221
Trainable params: 221
Non-trainable params: 0
_________________________________________________________________
acc: 77.08%

这篇关于在Keras中保存模型的正确方法的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！