儲存和讀取 TensorFlow 模型

• 儲存變數
• 載入變數
• 訓練一個模型並儲存它的權重
• 載入訓練好的模型

幸運的是，TensorFlow 可以讓你通過一個叫 tf.train.Saver 的類把你的程序儲存下來。這個類可以把任何 tf.Variable 存到你的檔案系統。

儲存變數

讓我們通過一個簡單地例子來儲存 weights 和 bias Tensors。第一個例子你只是存兩個變數，後面會教你如何把一個實際模型的所有權重儲存下來。

import tensorflow as tf

# The file path to save the data
# 檔案儲存路徑
save_file = './model.ckpt'

# Two Tensor Variables: weights and bias
# 兩個 Tensor 變數：權重和偏置項
weights = tf.Variable(tf.truncated_normal([2, 3]))
bias = tf.Variable(tf.truncated_normal([3]))

# Class used to save and/or restore Tensor Variables
# 用來存取 Tensor 變數的類
saver = tf.train.Saver()

with tf.Session() as sess:
    # Initialize all the Variables
    # 初始化所有變數
    sess.run(tf.global_variables_initializer())

    # Show the values of weights and bias
   # 顯示變數和權重
    print('Weights:')
    print(sess.run(weights))
    print('Bias:')
    print(sess.run(bias))

    # Save the model
    # 儲存模型
    saver.save(sess, save_file)
 

Weights:

[[-0.97990924 1.03016174 0.74119264]

[-0.82581609 -0.07361362 -0.86653847]]

Bias:

[ 1.62978125 -0.37812829 0.64723819]

weights 和 bias Tensors。 用 tf.truncated_normal() 函式設定了隨機值。用 tf.train.Saver.save()函式把這些值被儲存在save_file 位置，命名為 "model.ckpt"，（".ckpt" 擴充套件名錶示"checkpoint"）。

如果你使用 TensorFlow 0.11.0RC1 或者更新的版本，還會生成一個包含了 TensorFlow graph 的檔案 "model.ckpt.meta"

載入變數

現在這些變數已經存好了，讓我們把它們載入到新模型裡。

# Remove the previous weights and bias
# 移除之前的權重和偏置項
tf.reset_default_graph()

# Two Variables: weights and bias
# 兩個變數：權重和偏置項
weights = tf.Variable(tf.truncated_normal([2, 3]))
bias = tf.Variable(tf.truncated_normal([3]))

# Class used to save and/or restore Tensor Variables
# 用來存取 Tensor 變數的類
saver = tf.train.Saver()

with tf.Session() as sess:
    # Load the weights and bias
    # 載入權重和偏置項
    saver.restore(sess, save_file)

    # Show the values of weights and bias
    # 顯示權重和偏置項
    print('Weight:')
    print(sess.run(weights))
    print('Bias:')
    print(sess.run(bias))

輸出結果為：

    Weights:

    [[-0.97990924 1.03016174 0.74119264]

    [-0.82581609 -0.07361362 -0.86653847]]

    Bias:

    [ 1.62978125 -0.37812829 0.64723819]

注意，你依然需要在 Python 中建立 weights 和 bias Tensors。tf.train.Saver.restore() 函式把之前儲存的資料載入到 weights 和 bias 當中。

因為 tf.train.Saver.restore() 設定了 TensorFlow 變數，這裡你不需要呼叫 tf.global_variables_initializer()了。
儲存一個訓練好的模型

讓我們看看如何訓練一個模型並儲存它的權重。

訓練一個模型並儲存它的權重

從一個模型開始：

# Remove previous Tensors and Operations
# 移除之前的  Tensors 和運算
tf.reset_default_graph()

from tensorflow.examples.tutorials.mnist import input_data
import numpy as np

learning_rate = 0.001
n_input = 784  # MNIST 資料輸入 (圖片尺寸: 28*28)
n_classes = 10  # MNIST 總計類別 (數字 0-9)

# Import MNIST data
# 載入 MNIST 資料
mnist = input_data.read_data_sets('.', one_hot=True)

# Features and Labels
# 特徵和標籤
features = tf.placeholder(tf.float32, [None, n_input])
labels = tf.placeholder(tf.float32, [None, n_classes])

# Weights & bias
# 權重和偏置項
weights = tf.Variable(tf.random_normal([n_input, n_classes]))
bias = tf.Variable(tf.random_normal([n_classes]))

# Logits - xW + b
logits = tf.add(tf.matmul(features, weights), bias)

# Define loss and optimizer
# 定義損失函式和優化器
cost = tf.reduce_mean(\
    tf.nn.softmax_cross_entropy_with_logits(logits=logits, labels=labels))
optimizer = tf.train.GradientDescentOptimizer(learning_rate=learning_rate)\
    .minimize(cost)

# Calculate accuracy
# 計算準確率
correct_prediction = tf.equal(tf.argmax(logits, 1), tf.argmax(labels, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

讓我們訓練模型並儲存權重：

import math

save_file = './train_model.ckpt'
batch_size = 128
n_epochs = 100

saver = tf.train.Saver()

# Launch the graph
# 啟動圖
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())

    # Training cycle
    # 訓練迴圈
    for epoch in range(n_epochs):
        total_batch = math.ceil(mnist.train.num_examples / batch_size)

        # Loop over all batches
        # 遍歷所有 batch
        for i in range(total_batch):
            batch_features, batch_labels = mnist.train.next_batch(batch_size)
            sess.run(
                optimizer,
                feed_dict={features: batch_features, labels: batch_labels})

        # Print status for every 10 epochs
        # 每執行10個 epoch 列印一次狀態
        if epoch % 10 == 0:
            valid_accuracy = sess.run(
                accuracy,
                feed_dict={
                    features: mnist.validation.images,
                    labels: mnist.validation.labels})
            print('Epoch {:<3} - Validation Accuracy: {}'.format(
                epoch,
                valid_accuracy))

    # Save the model
    # 儲存模型
    saver.save(sess, save_file)
    print('Trained Model Saved.')

Epoch 0 - Validation Accuracy: 0.06859999895095825

Epoch 10 - Validation Accuracy: 0.20239999890327454

Epoch 20 - Validation Accuracy: 0.36980000138282776

Epoch 30 - Validation Accuracy: 0.48820000886917114

Epoch 40 - Validation Accuracy: 0.5601999759674072

Epoch 50 - Validation Accuracy: 0.6097999811172485

Epoch 60 - Validation Accuracy: 0.6425999999046326

Epoch 70 - Validation Accuracy: 0.6733999848365784

Epoch 80 - Validation Accuracy: 0.6916000247001648

Epoch 90 - Validation Accuracy: 0.7113999724388123

Trained Model Saved.

載入訓練好的模型

讓我們從磁碟中載入權重和偏置項，驗證測試集準確率。

saver = tf.train.Saver()

# Launch the graph
# 載入圖
with tf.Session() as sess:
    saver.restore(sess, save_file)

    test_accuracy = sess.run(
        accuracy,
        feed_dict={features: mnist.test.images, labels: mnist.test.labels})

print('Test Accuracy: {}'.format(test_accuracy))

Test Accuracy: 0.7229999899864197

就是這樣！你現在知道如何儲存再載入一個 TensorFlow 的訓練模型了。下一章節讓我們看看如何把權重和偏置項載入到修改過的模型中。