当前位置:网站首页>tf.GraphKeys
tf.GraphKeys
2022-07-26 19:12:00 【Wanderer001】
用于图形集合的标准名称。标准库使用各种名称来收集和检索与图关联的值。比如tf.Optimizer子类默认优化tf.GraphKeys下收集的变量。如果没有指定TRAINABLE_VARIABLES,也可以传递一个显式的变量列表。
定义了以下标准键:
- GLOBAL_VARIABLES:默认的变量对象集合,在分布式环境中共享(模型变量是这些对象的子集)。
- LOCAL_VARIABLES:每台机器的局部变量对象的子集,通常用于临时变量。
- MODEL_VARIABLES:模型中用于推理(前馈)的变量对象的子集。
- TRAINABLE_VARIABLES:优化器将训练的变量对象的子集。
QUEUE_RUNNERS:QueueRunner对象,用于为计算生成输入。- MOVING_AVERAGE_VARIABLES:也将保持移动平均线的变量对象的子集。
REGULARIZATION_LOSSES:在构建图期间收集的正则化损失。- SUMMARIES:日志生成相关的张量。
定义了以下标准键,但它们的集合不像其他键那样自动填充:
WEIGHTSBIASESACTIVATIONS
类成员:
ACTIVATIONS = 'activations'ASSET_FILEPATHS = 'asset_filepaths'BIASES = 'biases'CONCATENATED_VARIABLES = 'concatenated_variables'COND_CONTEXT = 'cond_context'EVAL_STEP = 'eval_step'GLOBAL_STEP = 'global_step'GLOBAL_VARIABLES = 'variables'INIT_OP = 'init_op'LOCAL_INIT_OP = 'local_init_op'LOCAL_RESOURCES = 'local_resources'LOCAL_VARIABLES = 'local_variables'LOSSES = 'losses'METRIC_VARIABLES = 'metric_variables'MODEL_VARIABLES = 'model_variables'MOVING_AVERAGE_VARIABLES = 'moving_average_variables'QUEUE_RUNNERS = 'queue_runners'READY_FOR_LOCAL_INIT_OP = 'ready_for_local_init_op'READY_OP = 'ready_op'REGULARIZATION_LOSSES = 'regularization_losses'RESOURCES = 'resources'SAVEABLE_OBJECTS = 'saveable_objects'SAVERS = 'savers'SUMMARIES = 'summaries'SUMMARY_OP = 'summary_op'TABLE_INITIALIZERS = 'table_initializer'TRAINABLE_RESOURCE_VARIABLES = 'trainable_resource_variables'TRAINABLE_VARIABLES = 'trainable_variables'TRAIN_OP = 'train_op'UPDATE_OPS = 'update_ops'这是一个tensorflow的计算图中内置的一个集合,其中会保存一些需要在训练操作之前完成的操作,并配合tf.control_dependencies函数使用。VARIABLES = 'variables'WEIGHTS = 'weights'WHILE_CONTEXT = 'while_context'
边栏推荐
- 3万脱发人,撑起一个IPO
- 金仓数据库 KingbaseES SQL 语言参考手册 (17. SQL语句: DISCARD 到 DROP LANGUAGE)
- ShardingSphere-JDBC 关键字问题
- 计算机组成原理常见面试题目总结,含答案
- 猎聘问卷星,成为微信「寄生虫」
- What should we do about the fragmentation of internal information? Try this
- 金仓数据库 KingbaseES SQL 语言参考手册 (20. SQL语句: MERGE 到 VALUES)
- Summary of iPhone development data persistence (final) - Comparison of five data persistence methods
- IM即时通讯开发如何压缩移动网络下APP的流量消耗
- iPhone开发 数据持久化总结(终结篇)—5种数据持久化方法对比
猜你喜欢

What should we do about the fragmentation of internal information? Try this

Server memory failure prediction can actually do this

Decompile jar files (idea environment)

Household deposits increased by 10.33 trillion yuan in the first half of the year, with an average of 57.1 billion deposits pouring into banks every day

Use of load balancing

靠元宇宙和NFT,天下秀疯狂“割韭菜”?

Leetcode daily practice - 88. Merge two ordered arrays

Training embedded representation of large categories using triple loss and twin neural networks

直播预约有奖| 高级咨询顾问徐雁斐:效能度量如何助力高效精细的外包管理

KVM virtualization
随机推荐
计算机专业面试题目总结,总导航
Design of intelligent weighing system based on Huawei cloud IOT (STM32) [i]
Linux regularly backs up the database and deletes the data n days ago
go+mysql+redis+vue3简单聊室,第5弹:使用消息队列和定时任务同步消息到mysql
6种方法帮你搞定SimpleDateFormat类不是线程安全的问题
IM即时通讯开发如何压缩移动网络下APP的流量消耗
The authentication type 10 is not supported
Deeply analyze the execution process of worker threads in the thread pool through the source code
Kingbases SQL language reference manual of Jincang database (15. SQL statement: create materialized view to create schema)
启牛是证券公司的渠道吗开户安全吗?
Leetcode daily practice - 26. Delete duplicates in an ordered array
Leetcode daily practice - 189. Rotation array
金仓数据库 KingbaseES SQL 语言参考手册 (13. SQL语句:ALTER SYNONYM 到 COMMENT)
【机器学习】变量间的相关性分析
Docker使用mysql:5.6和 owncloud 镜像,构建一个个人网盘,安装搭建私有仓库 Harbor
LeetCode每日一练 —— 26. 删除有序数组中的重复项
Collection of original IOS interview questions
【PHP】MySQL原生PHP操作-天龙八步
MySQL subquery usage
openstack 虚拟机网卡被重名为cirename0