NLU training issue

Rasa Open Source
1

I am making a bot.I needed some data for intents like greet,bye ,ask_whoisit ,canthelp ,ask_isbot ,deny ,affirm ,thank ,next_step etc just like your SARA… So I have taken some relevant data for me from SARA BOT git link.

I have made two bots with same data one with spacy and another with tensorflow pipeline.To see the difference.As of now I dont have any prior data but still wanted to see both’s work.

Tensor BOT:

nlu_config.yml:

language: en

pipeline: tensorflow_embedding

Spacy Bot:

nlu_config.yml:

language: en

pipeline: “spacy_sklearn”

The biggest wonder for me is , they both return me

{“intent”: { “name”: null, “confidence”: 1 }}

in nlu server when I parse any of the messages – thanks, see ya, catch you later, stop and more.

I have defined these exact words and sentences in my data.md file like below.

  {
    "intent": "canthelp",
    "text": "stop"
  },

  {
    "intent": "thank",
    "text": "thanks"
  },
  {
    "intent": "bye",
    "text": "see ya"
  },
  {
    "intent": "bye",
    "text": "catch you later"
  },

I believe the bot should recognize but its not.Please let me know where I have gone wrong. And other issue Iam confused is when I parse “bye”, it gives me intent name as goodbye as below json, but my intent was just “bye” as above.

    "intent": {
        "name": "goodbye",
        "confidence": 1
    },
    "entities": [],
    "text": "bye",
    "project": "default",
    "model": "fallback"
}

Please let me know where Iam going worng and suggest me any improvements.

Note that as of now I have not written any stories nor trained the rasa core.I first wanted to make sure my nlu is working okay.

2

我是用了tensortflow 和jieba_mitie_sklearn训练了2个机器人，在语料缺少的情况下，测试效果都不是很理想，你用的spacy，效果怎么样呢？

1 Like