Skip to content

CHANGELOG.md

Latest commit

 

History

History
255 lines (208 loc) · 7.64 KB

CHANGELOG.md

File metadata and controls

255 lines (208 loc) · 7.64 KB
Raw

Unreleased

1.4.1 - 2023-05-14

Added

  • BertEncoder输出使用mean输出而不再是cls输出

1.4.0 - 2022-11-26

Added

  • 添加繁体-简体互相转换功能 #11
  • 添加中文词列表合并字符串功能 #13

1.3.0 - 2022-11-26

Added

  • 添加macroF1评估函数 #12

1.2.2 - 2022-11-26

Fixed

  • 修复The truth value of a DataFrame is ambiguous #9

1.2.1 - 2021-11-22

Fixed

  • 确保代码兼容python3.6

1.2.0 - 2021-10-18

Added

  • 添加BERT预训练模型, 基于tensorflow hub
  • 添加针对tensor和ragged tensor的通用map函数
  • 实现一个函数导出所有的bert配置

1.1.4 - 2021-09-24

Fixed

  • 修复依赖文件
  • 修复log的继承关系
  • 修复语法上的问题

1.1.3 - 2021-09-24

Fixed

  • 修复缺少sklearn依赖的问题

1.1.2 - 2021-08-17

Added

  • 添加常用的自定义loss
  • 添加Logging的上下文管理模块, 控制节点
  • 添加字符串转驼峰脚本
  • 默认创建的logging添加stream logging功能

1.1.1 - 2021-07-18

Fixed

  • 修复Optional的错误
  • 默认的logger不允许从父类中继承handler
  • 修复分割数据集时, 大于1浮点数问题
  • 更新logging的默认名字
  • 更新发布流程: 打上标签即发布

1.1.0 - 2021-07-14

Added

  • 添加Bert token方法
  • 添加yaml文件读取工具
  • 添加keras模型快速编译函数
  • 添加keras模型基于dataset的训练脚本

Fixed

  • 修复logging模块重复输出日志的bug

1.0.0 - 2021-04-27

Added

  • 按照模块功能, 重新调整模型的结构, 调整后的模块为:
    • data: 数据预处理的各种功能函数
    • models: 自定义层, 常用模型
    • utils: 全局通用函数
    • metrics: 自定义衡量函数
    • pipeline: 流程控制函数
    • train: 自定义训练函数

0.9.0 - 2020-12-13

Added

  • Add setting GPU function
  • Add FernSimpleTrainer

Changed

  • Update trainer's annotation

0.8.1 - 2020-11-16

Added

  • Support parallel_apply for FernDataFrame
  • Update FernCleaner to process data with parallel process
  • Add test_parallel_apply method to test parallel_apply

Fixed

  • Map function error without send a function

0.8.0 - 2020-11-16

Added

  • Support parallel_map for FernSeries
  • Use Google annotation style in data.py

0.7.0 - 2020-11-11

Added

  • Support multi inputs and outputs with dictionary format data
  • Label weight is no longer recommended, but dataset balancing is
  • Use a stub file to annotate the parameter types of the file train.py

0.6.4 - 2020-11-10

Fixed

  • Fix data set length error when there is a multi input dictionary data

0.6.3 - 2020-10-16

Added

  • ZeroDivisionError, while balancing data and num = 1

0.6.2 - 2020-10-15

Fixed

  • Prefix code type error
  • Data and label column name error while training
  • Data type error while loading data from disk
  • Repeat prefix code

0.6.1 - 2020-10-11

Fixed

  • No longer to delete character '<' and '>' for keeping special words like

0.6.0 - 2020-10-11

Added

  • Let FernDownloader download data from sql via SQLAlchemy
  • Treat a string like as a word not a string

Removed

  • Delete all unimportant words

Fixed

  • KeyError while concatenate a Series to an array

0.5.1 - 2020-07-27

Fixed

  • Fix dataset_total length

Changed

  • Change loss function's train param to with_label_weight

0.5.0 - 2020-07-26

Added

  • Add FernBalance class for dataset balance
  • Add gpu finder to find which gpu has the largest free memory

0.4.1 - 2020-07-22

Added

  • Stop words use regular matching
  • The # comment rule is allowed in word library

0.4.0 - 2020-07-21

Added

  • Add optimized TextCNN model as a built-in model

0.3.3 - 2020-07-09

Added

  • Add predict function for FernModel

0.3.2 - 2020-07-06

Changed

  • Rename BaseTrainer -> FernTrainer

Fixed

  • Fix trainer not fitting for multi output labels

0.3.1 - 2020-07-02

Fixed

  • Fix no user words file exist issue
  • Fix cleaned label value should be a list issue

0.3.0 - 2020-07-02

Added

  • using cut_func to split sequence to word list
  • Add save_function for every data precessing stage
  • Mark FernTransformer.output_shape as optional
  • Enable data frame default index
  • Add Sequence2Words class as cut_func generator
  • Use fern as default data frame

0.2.0 - 2020-05-24

Added

  • Add label path for storing label data

Changed

  • Modify output shape from list into dict
  • Modify transformed label from array into dictionary
  • Update transformed label and data function
  • Change variable name input_col to data_col
  • Change variable name output_col to label_col

0.1.4 - 2020-05-20

Changed

  • Change the random state location where it's defined

0.1.3 - 2020-05-20

Added

  • add random state for data splitter

0.1.2 - 2020-05-07

Fixed

  • Fix training loop quick exit bug

0.1.1 - 2020-05-03

Added

  • Add github python publish action

0.1.0 - 2020-05-03

Added

  • Data preprocessing module: data downloader, data cleaner, data transformer and data splitter
  • Model template: model builder, saver, loader and structure printer
  • Model trainer
  • Custom layers: Conv1DPassMask, FlattenPassMask, DenseWithMask, AttentionLayer and ScaledDotProductAttention
  • Custom metrics: BinaryCategoricalAccuarcy
  • Other tools: logging and progress bar
  • Add README
  • Add README_ZH