kendricktan / laice

Train your own Natural Language Processor from a browser 🤖 (Prototype)
MIT License
172 stars 27 forks source link

Response wrongly formatted #6

Open alekaizer opened 7 years ago

alekaizer commented 7 years ago

Hi guys,

I just got the app up and running and from my first tentative, the output is wrongly formatted.

Here is what I was supposed to get:

the output keys are at the right places, but on my output, the keys are the values, and it doesn't get rid of punctuations.

[
   {
      "id":6,
      "querystring":"where can I get a car",
      "configured":false,
      "parsed_ner":{
         "get":"action",
         "where":"object"
      }
   },
   {
      "id":7,
      "querystring":"how to drive a car",
      "configured":false,
      "parsed_ner":{
         "car":"object",
         "drive":"action"
      }
   },
   {
      "id":8,
      "querystring":"can you teach me how to shoot a basketball?",
      "configured":false,
      "parsed_ner":{
         "me":"action",
         "teach":"action",
         "shoot":"action",
         "basketball":"object",
         "?":"action"
      }
   }
]

for the weird parts, me,where taken as valid in parsing, I guess it's because of the tiny size of the training dataset

kendricktan commented 7 years ago

a general rule of thumb is that you would need around 5-10 training datasets before it'll achieve the intended result