Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/hscspring/chatbot
Lab for Chatbot
https://github.com/hscspring/chatbot
Last synced: 7 days ago
JSON representation
Lab for Chatbot
- Host: GitHub
- URL: https://github.com/hscspring/chatbot
- Owner: hscspring
- License: apache-2.0
- Created: 2021-09-30T12:34:35.000Z (about 3 years ago)
- Default Branch: main
- Last Pushed: 2021-09-30T14:09:48.000Z (about 3 years ago)
- Last Synced: 2024-10-18T00:46:37.501Z (26 days ago)
- Language: Python
- Size: 986 KB
- Stars: 0
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# README
## 实验步骤
### 了解数据
通过阅读论文《MultiWOZ 2.1: A Consolidated Multi-Domain Dialogue Dataset with State Corrections and State Tracking Baselines》了解数据基本结构。
### 了解评估方法
通过阅读论文《ConvLab-2: An Open-Source Toolkit for Building, Evaluating, and Diagnosing Dialogue Systems》了解如何评估一个对话系统。
### 重构评估 Agent
ConvLab-2 使用 sys_content 训练数据训练一个模拟器作为 user 角色对对话系统能力进行模拟评估。通过阅读源代码，将这部分内容单独重构为一个模块，使用方法如下：
```python
user_nlu = BERTNLU(
model_dir=os.path.join(root, "model/sys_context"),
vocab_dir=os.path.join(root, "data/agent/vocab"),
)
user_dst = None
user_policy = RulePolicy(
goal_model_path=os.path.join(root, "model/goal/new_goal_model.pkl"),
db_path=os.path.join(root, "data/agent/db"),
vocab_path=os.path.join(root, "data/agent/vocab/"),
character="usr",
)
user_nlg = TemplateNLG(
is_user=True,
template_dir=os.path.join(root, "data/agent/template")
)
user_agent = PipelineAgent(user_nlu, user_dst, user_policy, user_nlg, name='user')
analyzer = Analyzer(
db_path=os.path.join(root, "data/agent/db"),
user_agent=user_agent,
dataset='multiwoz'
)
```
运行 `damd` 作为 System Agent，以 100 次对话为例：
```python
sys_agent = Damd(
vocab_path=os.path.join(root, "data/damd/vocab"),
data_path=os.path.join(root, "data/damd/data_processed"),
db_processed_path=os.path.join(root, "data/damd/db_processed"),
model_path=os.path.join(root, "model/damd/all_aug_sample3_sd777_lr0.005_bs80_sp5_dc3")
)
analyzer.sample_dialog(sys_agent)
analyzer.comprehensive_analyze(sys_agent=sys_agent, model_name='DAMD', total_dialog=100)
```
### 阅读论文
分别阅读论文，了解思想和算法。
- SimpleTOD: A Simple Language Model for Task-Oriented Dialogue：使用 GPT2 训练端到端模型
- UBAR: Towards Fully End-to-End Task-Oriented Dialog System with GPT-2：相比 SImpleTOD，上下文增加考虑了 Belief State，DataBase Result，System Action
### 训练模型
需要将原来的 2.0 版本数据调整为 2.1 版本。
#### simpletod
训练代码：
```bash
$ zsh train_end2end.sh cpu gpt2 gpt2-tiny 2
```
#### ubar
训练代码：
```bash
$ python3 train.py -mode train -cfg gpt_path=distilgpt2 lr=1e-4 warmup_steps=2000 gradient_accumulation_steps=16 batch_size=2 epoch_num=60 exp_no=best_model
```
### 评估推理
使用之前构建好的 `chatbot_agent` 作为 user，新的模型作为 system（可参考之前的 damd），完成自动交互评估测试。
#### simpletod
```python
sys_agent = SimpleTod(
data_path = os.path.join(root, "data/simpletod"),
model_path = os.path.join(root, "model/simpletod/gpt2-small/checkpoint-111")
)
analyzer.sample_dialog(sys_agent)
analyzer.comprehensive_analyze(sys_agent=sys_agent, model_name='simpletod', total_dialog=10)
```
#### ubar
todo
### 使用指南
依赖：Python3.7
```bash
$ pip install -r requirements.txt
$ cd chatbot_agent
$ pip install -e .
$ cd e2e_dialog
$ pip install -e .
```
下载 data 和 model 到根目录：
- 百度盘：链接: https://pan.baidu.com/s/13vIXu59iT5SdSKVFyAzqrw 提取码: 9d22
解压后如下所示：
```bash
├── chatbot_agent
├── data
│ ├── agent
│ ├── damd
│ ├── simpletod
│ └── ubar
├── e2e_dialog
├── eval
├── model
│ ├── damd
│ ├── goal
│ ├── simpletod
│ ├── sys_context
│ └── ubar
├── requirements.txt
```
下载 Spacy 模型：
```bash
$ python -m spacy download en_core_web_sm
```
在 eval 目录下执行对应的模型：
```bash
$ python simpletod.py
$ python damd.py
```
## 附录
### 项目整体目录
```bash
├── chatbot
├── damd
├── simpletod
└── ubar
```
### chatbot 目录
```bash
.
├── README.md
├── chatbot_agent
│ ├── README.md
│ ├── chatbot_agent
│ └── setup.py
├── data
│ ├── agent
│ ├── damd
│ ├── simpletod
│ └── ubar
├── e2e_dialog
│ ├── README.md
│ ├── damd
│ ├── setup.py
│ ├── simpletod
│ └── ubar
├── eval
│ ├── __pycache__
│ ├── damd.py
│ ├── results
│ ├── simpletod.py
│ ├── ubar.py
│ ├── user.py
│ └── user.pyc
├── images
│ ├── damd.png
│ ├── simpletod.png
│ └── ubar.png
├── model
│ ├── damd
│ ├── goal
│ ├── simpletod
│ ├── sys_context
│ └── ubar
└── requirements.txt
```