Rasa入门指南

1. Rasa是什么？

Rasa是一个开源机器学习框架，用于构建语境化AI助手和聊天机器人。

Rasa有两个主要的模块：

• NLU 用于理解用户的消息
• Core 用于管理对话并且决定下一步做什么

RasaX是一个帮你构建、开发、部署AI助手的工具借助Rasa框架，RasaX包含一个用户接口和一个REST API。

Rasa+Rasa X

2.使用Rasa构建AI助手

rasa init命令创建Rasa项目脚手架。这包括培训数据和一些配置文件。它还将使用一些示例训练数据训练您的第一个模型。

rasa init --no-prompt

目录结构：

项目目录

# 执行结果

Welcome to Rasa! 🤖

To get started quickly, an initial project will be created.
If you need some help, check out the documentation at https://rasa.com/docs/rasa.

Created project directory at '/Users/weipengfei/workspaces/RasaProjects'.
Finished creating project structure.
Training an initial model...
Training Core model...
Processed Story Blocks: 100%|█| 4/4 [00:00<00:00, 4236.67it/s, # trackers=1]
Processed Story Blocks: 100%|█| 4/4 [00:00<00:00, 2035.58it/s, # trackers=4]
Processed Story Blocks: 100%|█| 4/4 [00:00<00:00, 641.23it/s, # trackers=12]
Processed Story Blocks: 100%|██| 4/4 [00:00<00:00, 934.09it/s, # trackers=7]
Processed trackers: 100%|█████| 4/4 [00:00<00:00, 2861.05it/s, # actions=14]
Processed actions: 14it [00:00, 8074.84it/s, # examples=14]
Processed trackers: 100%|███| 94/94 [00:00<00:00, 1608.27it/s, # actions=62]
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
masking (Masking)            (None, 5, 19)             0         
_________________________________________________________________
lstm (LSTM)                  (None, 32)                6656      
_________________________________________________________________
dense (Dense)                (None, 13)                429       
_________________________________________________________________
activation (Activation)      (None, 13)                0         
=================================================================
Total params: 7,085
Trainable params: 7,085
Non-trainable params: 0
_________________________________________________________________
2019-07-29 11:10:29 INFO     rasa.core.policies.keras_policy  - Fitting model with 62 total samples and a validation split of 0.1
Epoch 1/100
62/62 [==============================] - 1s 12ms/sample - loss: 2.5240 - acc: 0.2903
Epoch 2/100
62/62 [==============================] - 0s 188us/sample - loss: 2.4984 - acc: 0.3226
Epoch 3/100
62/62 [==============================] - 0s 194us/sample - loss: 2.4557 - acc: 0.3548
Epoch 4/100
62/62 [==============================] - 0s 199us/sample - loss: 2.4287 - acc: 0.3548
Epoch 5/100
62/62 [==============================] - 0s 247us/sample - loss: 2.4104 - acc: 0.4194
Epoch 6/100
62/62 [==============================] - 0s 192us/sample - loss: 2.3961 - acc: 0.4032
Epoch 7/100
62/62 [==============================] - 0s 205us/sample - loss: 2.3686 - acc: 0.4355
Epoch 8/100
62/62 [==============================] - 0s 186us/sample - loss: 2.3098 - acc: 0.4194
Epoch 9/100
62/62 [==============================] - 0s 194us/sample - loss: 2.2944 - acc: 0.4194
Epoch 10/100
62/62 [==============================] - 0s 197us/sample - loss: 2.2666 - acc: 0.4516
Epoch 11/100
62/62 [==============================] - 0s 186us/sample - loss: 2.2202 - acc: 0.4355
Epoch 12/100
62/62 [==============================] - 0s 200us/sample - loss: 2.2005 - acc: 0.4355
Epoch 13/100
62/62 [==============================] - 0s 186us/sample - loss: 2.1775 - acc: 0.4355
Epoch 14/100
62/62 [==============================] - 0s 204us/sample - loss: 2.1157 - acc: 0.4355
Epoch 15/100
62/62 [==============================] - 0s 187us/sample - loss: 2.0968 - acc: 0.4355
Epoch 16/100
62/62 [==============================] - 0s 221us/sample - loss: 2.0557 - acc: 0.4355
Epoch 17/100
62/62 [==============================] - 0s 188us/sample - loss: 2.0413 - acc: 0.4355
Epoch 18/100
62/62 [==============================] - 0s 186us/sample - loss: 2.0061 - acc: 0.4355
Epoch 19/100
62/62 [==============================] - 0s 191us/sample - loss: 1.9796 - acc: 0.4355
Epoch 20/100
62/62 [==============================] - 0s 180us/sample - loss: 1.9174 - acc: 0.4355
Epoch 21/100
62/62 [==============================] - 0s 197us/sample - loss: 1.9245 - acc: 0.4355
Epoch 22/100
62/62 [==============================] - 0s 182us/sample - loss: 1.8879 - acc: 0.4355
Epoch 23/100
62/62 [==============================] - 0s 220us/sample - loss: 1.8528 - acc: 0.4355
Epoch 24/100
62/62 [==============================] - 0s 182us/sample - loss: 1.7882 - acc: 0.4355
Epoch 25/100
62/62 [==============================] - 0s 189us/sample - loss: 1.8093 - acc: 0.4355
Epoch 26/100
62/62 [==============================] - 0s 181us/sample - loss: 1.7950 - acc: 0.4355
Epoch 27/100
62/62 [==============================] - 0s 193us/sample - loss: 1.7785 - acc: 0.4355
Epoch 28/100
62/62 [==============================] - 0s 197us/sample - loss: 1.7653 - acc: 0.4355
Epoch 29/100
62/62 [==============================] - 0s 221us/sample - loss: 1.7613 - acc: 0.4355
Epoch 30/100
62/62 [==============================] - 0s 216us/sample - loss: 1.7423 - acc: 0.4355
Epoch 31/100
62/62 [==============================] - 0s 185us/sample - loss: 1.7363 - acc: 0.4355
Epoch 32/100
62/62 [==============================] - 0s 238us/sample - loss: 1.6864 - acc: 0.4355
Epoch 33/100
62/62 [==============================] - 0s 198us/sample - loss: 1.6771 - acc: 0.4355
Epoch 34/100
62/62 [==============================] - 0s 191us/sample - loss: 1.6826 - acc: 0.4355
Epoch 35/100
62/62 [==============================] - 0s 180us/sample - loss: 1.6655 - acc: 0.4355
Epoch 36/100
62/62 [==============================] - 0s 190us/sample - loss: 1.6314 - acc: 0.4355
Epoch 37/100
62/62 [==============================] - 0s 183us/sample - loss: 1.6295 - acc: 0.4355
Epoch 38/100
62/62 [==============================] - 0s 181us/sample - loss: 1.5983 - acc: 0.4355
Epoch 39/100
62/62 [==============================] - 0s 207us/sample - loss: 1.6012 - acc: 0.4355
Epoch 40/100
62/62 [==============================] - 0s 230us/sample - loss: 1.5771 - acc: 0.4355
Epoch 41/100
62/62 [==============================] - 0s 227us/sample - loss: 1.5864 - acc: 0.4355
Epoch 42/100
62/62 [==============================] - 0s 214us/sample - loss: 1.5663 - acc: 0.4355
Epoch 43/100
62/62 [==============================] - 0s 227us/sample - loss: 1.5479 - acc: 0.4355
Epoch 44/100
62/62 [==============================] - 0s 244us/sample - loss: 1.5487 - acc: 0.4355
Epoch 45/100
62/62 [==============================] - 0s 237us/sample - loss: 1.5426 - acc: 0.4355
Epoch 46/100
62/62 [==============================] - 0s 243us/sample - loss: 1.5315 - acc: 0.4355
Epoch 47/100
62/62 [==============================] - 0s 246us/sample - loss: 1.5264 - acc: 0.4355
Epoch 48/100
62/62 [==============================] - 0s 241us/sample - loss: 1.5074 - acc: 0.4355
Epoch 49/100
62/62 [==============================] - 0s 241us/sample - loss: 1.5014 - acc: 0.4355
Epoch 50/100
62/62 [==============================] - 0s 212us/sample - loss: 1.4918 - acc: 0.4355
Epoch 51/100
62/62 [==============================] - 0s 234us/sample - loss: 1.5033 - acc: 0.4355
Epoch 52/100
62/62 [==============================] - 0s 233us/sample - loss: 1.4698 - acc: 0.4355
Epoch 53/100
62/62 [==============================] - 0s 216us/sample - loss: 1.4486 - acc: 0.4355
Epoch 54/100
62/62 [==============================] - 0s 277us/sample - loss: 1.4537 - acc: 0.4355
Epoch 55/100
62/62 [==============================] - 0s 214us/sample - loss: 1.4533 - acc: 0.4355
Epoch 56/100
62/62 [==============================] - 0s 224us/sample - loss: 1.4438 - acc: 0.4355
Epoch 57/100
62/62 [==============================] - 0s 252us/sample - loss: 1.4295 - acc: 0.4355
Epoch 58/100
62/62 [==============================] - 0s 289us/sample - loss: 1.4214 - acc: 0.4355
Epoch 59/100
62/62 [==============================] - 0s 247us/sample - loss: 1.4170 - acc: 0.4355
Epoch 60/100
62/62 [==============================] - 0s 213us/sample - loss: 1.4095 - acc: 0.4355
Epoch 61/100
62/62 [==============================] - 0s 302us/sample - loss: 1.3916 - acc: 0.4355
Epoch 62/100
62/62 [==============================] - 0s 232us/sample - loss: 1.3877 - acc: 0.4355
Epoch 63/100
62/62 [==============================] - 0s 228us/sample - loss: 1.3765 - acc: 0.4355
Epoch 64/100
62/62 [==============================] - 0s 299us/sample - loss: 1.3811 - acc: 0.4355
Epoch 65/100
62/62 [==============================] - 0s 256us/sample - loss: 1.3795 - acc: 0.4355
Epoch 66/100
62/62 [==============================] - 0s 291us/sample - loss: 1.3574 - acc: 0.4355
Epoch 67/100
62/62 [==============================] - 0s 254us/sample - loss: 1.3492 - acc: 0.4355
Epoch 68/100
62/62 [==============================] - 0s 209us/sample - loss: 1.3499 - acc: 0.4355
Epoch 69/100
62/62 [==============================] - 0s 209us/sample - loss: 1.3304 - acc: 0.4355
Epoch 70/100
62/62 [==============================] - 0s 215us/sample - loss: 1.3185 - acc: 0.4355
Epoch 71/100
62/62 [==============================] - 0s 228us/sample - loss: 1.3221 - acc: 0.4677
Epoch 72/100
62/62 [==============================] - 0s 261us/sample - loss: 1.3000 - acc: 0.4677
Epoch 73/100
62/62 [==============================] - 0s 209us/sample - loss: 1.2968 - acc: 0.4516
Epoch 74/100
62/62 [==============================] - 0s 225us/sample - loss: 1.3253 - acc: 0.4677
Epoch 75/100
62/62 [==============================] - 0s 234us/sample - loss: 1.2877 - acc: 0.4677
Epoch 76/100
62/62 [==============================] - 0s 202us/sample - loss: 1.2892 - acc: 0.4839
Epoch 77/100
62/62 [==============================] - 0s 221us/sample - loss: 1.2595 - acc: 0.4839
Epoch 78/100
62/62 [==============================] - 0s 200us/sample - loss: 1.2663 - acc: 0.4839
Epoch 79/100
62/62 [==============================] - 0s 221us/sample - loss: 1.2466 - acc: 0.5000
Epoch 80/100
62/62 [==============================] - 0s 216us/sample - loss: 1.2508 - acc: 0.4839
Epoch 81/100
62/62 [==============================] - 0s 189us/sample - loss: 1.2334 - acc: 0.4677
Epoch 82/100
62/62 [==============================] - 0s 223us/sample - loss: 1.2180 - acc: 0.4839
Epoch 83/100
62/62 [==============================] - 0s 227us/sample - loss: 1.2409 - acc: 0.4677
Epoch 84/100
62/62 [==============================] - 0s 218us/sample - loss: 1.2258 - acc: 0.4677
Epoch 85/100
62/62 [==============================] - 0s 249us/sample - loss: 1.1977 - acc: 0.4839
Epoch 86/100
62/62 [==============================] - 0s 210us/sample - loss: 1.2270 - acc: 0.4839
Epoch 87/100
62/62 [==============================] - 0s 217us/sample - loss: 1.2157 - acc: 0.5161
Epoch 88/100
62/62 [==============================] - 0s 216us/sample - loss: 1.1658 - acc: 0.5161
Epoch 89/100
62/62 [==============================] - 0s 202us/sample - loss: 1.1864 - acc: 0.4839
Epoch 90/100
62/62 [==============================] - 0s 205us/sample - loss: 1.1830 - acc: 0.5161
Epoch 91/100
62/62 [==============================] - 0s 216us/sample - loss: 1.1624 - acc: 0.5161
Epoch 92/100
62/62 [==============================] - 0s 192us/sample - loss: 1.1616 - acc: 0.5323
Epoch 93/100
62/62 [==============================] - 0s 212us/sample - loss: 1.1494 - acc: 0.5484
Epoch 94/100
62/62 [==============================] - 0s 189us/sample - loss: 1.1198 - acc: 0.5323
Epoch 95/100
62/62 [==============================] - 0s 210us/sample - loss: 1.1305 - acc: 0.5484
Epoch 96/100
62/62 [==============================] - 0s 222us/sample - loss: 1.1220 - acc: 0.5484
Epoch 97/100
62/62 [==============================] - 0s 234us/sample - loss: 1.0849 - acc: 0.5968
Epoch 98/100
62/62 [==============================] - 0s 198us/sample - loss: 1.1161 - acc: 0.5645
Epoch 99/100
62/62 [==============================] - 0s 195us/sample - loss: 1.1265 - acc: 0.5484
Epoch 100/100
62/62 [==============================] - 0s 216us/sample - loss: 1.1141 - acc: 0.5484
2019-07-29 11:10:32 INFO     rasa.core.policies.keras_policy  - Done fitting keras policy model
2019-07-29 11:10:33 INFO     rasa.core.agent  - Persisted model to '/var/folders/9y/xksbgbfx79sgdrf18b6vh6gw0000gn/T/tmplricej0h/core'
Core model training completed.
Training NLU model...
2019-07-29 11:10:33 INFO     rasa.nlu.training_data.loading  - Training data format of /var/folders/9y/xksbgbfx79sgdrf18b6vh6gw0000gn/T/tmpegquajbg/7fd2d266caa343cfa3332f6de40b69dd_nlu.md is md
2019-07-29 11:10:33 INFO     rasa.nlu.training_data.training_data  - Training data stats: 
    - intent examples: 39 (6 distinct intents)
    - Found intents: 'greet', 'mood_great', 'goodbye', 'deny', 'affirm', 'mood_unhappy'
    - entity examples: 0 (0 distinct entities)
    - found entities: 

2019-07-29 11:10:33 INFO     rasa.nlu.model  - Starting to train component WhitespaceTokenizer
2019-07-29 11:10:33 INFO     rasa.nlu.model  - Finished training component.
2019-07-29 11:10:33 INFO     rasa.nlu.model  - Starting to train component RegexFeaturizer
2019-07-29 11:10:33 INFO     rasa.nlu.model  - Finished training component.
2019-07-29 11:10:33 INFO     rasa.nlu.model  - Starting to train component CRFEntityExtractor
2019-07-29 11:10:33 INFO     rasa.nlu.model  - Finished training component.
2019-07-29 11:10:33 INFO     rasa.nlu.model  - Starting to train component EntitySynonymMapper
2019-07-29 11:10:33 INFO     rasa.nlu.model  - Finished training component.
2019-07-29 11:10:33 INFO     rasa.nlu.model  - Starting to train component CountVectorsFeaturizer
2019-07-29 11:10:33 INFO     rasa.nlu.model  - Finished training component.
2019-07-29 11:10:33 INFO     rasa.nlu.model  - Starting to train component EmbeddingIntentClassifier
2019-07-29 11:10:33 INFO     rasa.nlu.classifiers.embedding_intent_classifier  - Accuracy is updated every 10 epochs
Epochs: 100%|████████████████████████████████████████████████████████████████████████| 300/300 [00:01<00:00, 276.65it/s, loss=0.091, acc=1.000]
2019-07-29 11:10:34 INFO     rasa.nlu.classifiers.embedding_intent_classifier  - Finished training embedding classifier, loss=0.091, train accuracy=1.000
2019-07-29 11:10:34 INFO     rasa.nlu.model  - Finished training component.
2019-07-29 11:10:35 INFO     rasa.nlu.model  - Successfully saved model into '/var/folders/9y/xksbgbfx79sgdrf18b6vh6gw0000gn/T/tmplricej0h/nlu'
NLU model training completed.
Your Rasa model is trained and saved at '/Users/weipengfei/workspaces/RasaProjects/models/20190729-111027.tar.gz'.
If you want to speak to the assistant, run 'rasa shell' at any time inside the project directory.

运行shell：

rasa shell

# 结果
![rasa shell](https://img.haomeiwen.com/i4905462/3ce80ac380b34182.png?imageMogr2/auto-orient/strip%7CimageView2/2/w/1240)

3. 使用Rasa X学习真实对话

# 安装
pip install rasa-x --extra-index-url https://pypi.rasa.com/simple
# 执行
cd RasaProjects & rasa x

效果：

Rasa X启动后效果图