TensorFlow Saver 保存最佳模型 tf.train.Saver Save Best Model

TensorFlow Saver 保存最佳模型 tf.train.Saver Save Best Model

Checkmate is designed to be a simple drop-in solution for a very common Tensorflow use-case: keeping track of the best model checkpoints during training.

The BestCheckpointSaver is a wrapper around a tf.train.Saver.

The BestCheckpointSaver provides the ability to save the best n checkpoints, whereas the tf.train.Saver can only save the last n checkpoints.

Features

  • Save only best n checkpoints
  • Compares checkpoints based on a user-provided value
  • Can rank checkpoints by highest or lowest values
  • Automatically delete outdated checkpoints
  • Provide at a glance record of each checkpoint‘s associated value (the user-provided value obtained from that checkpoint)

Using the BestCheckpointSaver

from checkmate import BestCheckpointSaver

# ...build model...

best_ckpt_saver = BestCheckpointSaver(
  save_dir=best_checkpoint_dir,
  num_to_keep=3,
  maximize=True
)

# train and evaluate
for train_step in range(max_steps):
  sess.run(train_op)
  if train_step % evaluation_interval == 0:
    accuracy = sess.run(eval_op, feed_dict=validation_data)
    best_ckpt_saver.handle(accuracy, sess, global_step_tensor)

Loading the best checkpoint

import checkmate

# ...build model...

saver = tf.train.Saver()
saver.restore(sess, checkmate.get_best_checkpoint(best_checkpoint_dir, maximize=True))

At this stage, the module is no-frills with limited documentation. It is not intended to work in distributed settings or with complex Session/Graph management (i.e. the tf.Estimator framework). Contributions are welcome.

来源： https://github.com/vonclites/checkmate

原文地址：https://www.cnblogs.com/jins-note/p/10246750.html

时间： 2024-10-12 20:26:04

TensorFlow Saver 保存最佳模型 tf.train.Saver Save Best Model的相关文章

跟我学算法- tensorflow模型的保存与读取 tf.train.Saver()

save =  tf.train.Saver() 通过save. save() 实现数据的加载 通过save.restore() 实现数据的导出 第一步: 数据的载入 import tensorflow as tf #创建变量 v1 = tf.Variable(tf.random_normal([1, 2], name='v1')) v2 = tf.Variable(tf.random_normal([2, 3], name='v2')) #初始化变量 init_op = tf.global_v

tf.train.Saver()-tensorflow中模型的保存及读取

作用:训练网络之后保存训练好的模型,以及在程序中读取已保存好的模型 使用步骤: 实例化一个Saver对象 saver = tf.train.Saver() 在训练过程中,定期调用saver.save方法,像文件夹中写入包含当前模型中所有可训练变量的checkpoint文件 saver.save(sess,FLAGG.train_dir,global_step=step) 之后可以使用saver.restore()方法,重载模型的参数,继续训练或者用于测试数据 saver.restore(sess

图融合之加载子图：Tensorflow.contrib.slim与tf.train.Saver之坑

import tensorflow as tf import tensorflow.contrib.slim as slim import rawpy import numpy as np import tensorflow as tf import struct import glob import os from PIL import Image import time __sony__ = 0 __huawei__ = 1 __blackberry__ = 2 __stage_raw2ra

机器学习与Tensorflow（7）——tf.train.Saver()、inception-v3的应用

1. tf.train.Saver() tf.train.Saver()是一个类,提供了变量.模型(也称图Graph)的保存和恢复模型方法. TensorFlow是通过构造Graph的方式进行深度学习,任何操作(如卷积.池化等)都需要operator,保存和恢复操作也不例外. 在tf.train.Saver()类初始化时,用于保存和恢复的save和restore operator会被加入Graph.所以,下列类初始化操作应在搭建Graph时完成. saver = tf.train.Saver()

tensorflow-训练检查点tf.train.Saver

#!/usr/bin/env python2 # -*- coding: utf-8 -*- """ Created on Thu Sep 6 10:16:37 2018 @author: myhaspl @email:[email protected] """ import tensorflow as tf g1=tf.Graph() with g1.as_default(): with tf.name_scope("input_Va

tf.train.Saver()

1. 实例化对象 saver = tf.train.Saver(max_to_keep=1) max_to_keep: 表明保存的最大checkpoint文件数.当一个新文件创建的时候,旧文件就会被删掉.如果值为None或0, 表示保存所有的checkpoint文件.默认值5(也就是说,保存最近的5个checkpoint文件). keep_checkpoint_every_n_hour: 除了保存最近的max_to_keep_checkpoint文件,你还可能想每训练N小时保存一个checkpo

TensorFlow 实战（二）—— tf train（优化算法）

Training | TensorFlow tf 下以大写字母开头的含义为名词的一般表示一个类(class) 1. 优化器(optimizer) 优化器的基类(Optimizer base class)主要实现了两个接口,一是计算损失函数的梯度,二是将梯度作用于变量.tf.train 主要提供了如下的优化函数: tf.train.Optimizer tf.train.GradientDescentOptimizer tf.train.AdadeltaOpzimizer Ada delta tf.

TF：利用TF的train.Saver将训练好的variables(W、b)保存到指定的index、meda文件

import tensorflow as tf import numpy as np W = tf.Variable([[2,1,8],[1,2,5]], dtype=tf.float32, name='weights') b = tf.Variable([[1,2,5]], dtype=tf.float32, name='biases') init= tf.global_variables_initializer() saver = tf.train.Saver() with tf.Sessi

TF：利用TF的train.Saver载入曾经训练好的variables(W、b)以供预测新的数据

import tensorflow as tf import numpy as np W = tf.Variable(np.arange(6).reshape((2, 3)), dtype=tf.float32, name="weights") b = tf.Variable(np.arange(3).reshape((1, 3)), dtype=tf.float32, name="biases") saver = tf.train.Saver() with tf.