Releases: shenweichen/DeepMatch
Releases · shenweichen/DeepMatch
v0.3.1
DeepMatch
Major Features and Improvements
Bug Fixes and Other Changes
- Update evaluation logic in multi interets model @shenweichen
DisscussionGroup
- Github Discussions
- Wechat Discussions
公众号:浅梦学习笔记 | 微信:deepctrbot | 学习小组 加入 主题集合 |
---|---|---|
v0.3.0
DeepMatch
⚠️ This version does not support backward compatibility
- this verson is not compatible with v0.2.* and v0.1.*
- 该版本不支持向下兼容,请按照样例进行完整的样本处理和模型构建
Major Features and Improvements
- Support different negative sampling strategies, including
inbatch
,uniform
,frequency
,adaptive
. @shenweichen - refactor & update
preprocess.py
, @shenweichen - DSSM inbatch negative sampling example
- YoutubeDNN/MIND frequency negtivte sampling example
- SDM frequency negtivte sampling example
API Changes
- Add new class
deepmatch.utils.NegativeSampler
- Remove
deepmatch.layers.core.Similarity
- Add
sampler_config
,loss_type
parameters for models - Change
gamma
totemperature
,gamma=1/temperature
DisscussionGroup
- Github Discussions
- Wechat Discussions
公众号:浅梦学习笔记 | 微信:deepctrbot | 学习小组 加入 主题集合 |
---|---|---|
v0.2.1
Deepmatch
Major Features and Improvements
- Add Scaling factor gamma in DSSM,FM @zzszmyf
- Update layers @shenweichen
Bug Fixes and Other Changes
- Fix routing_logits error of MIND #44 #57 #75 @clhchtcjj
DisscussionGroup
- Github Discussions
- Wechat Discussions
公众号:浅梦学习笔记 | 微信:deepctrbot | 学习小组 加入 主题集合 |
---|---|---|
v0.2.0
Change Log
Major Features and Improvements
- Support different initializers for different embedding weights and loading pretrained embeddings.
v0.1.3
v0.1.2
- support saving&loading model
- support higher version of tensorflow
- add Google colab notebook