Auto Train your wit.ai bot using YAML format dataset. The goal of making this program are:

• Avoid the task of submitting data one-by-one through the Web GUI
• Minimize the use of code when training your bot
• Make it easier to train using entities tagged data

This program runs on Python 2 or Python 3 and needs some requirements written in requirements.txt. Install it using the command below (after you install pip on your system):

pip install -r requirements.txt

For every bot you want to make, create a single folder which contains config file (name it _config.yaml) and dataset files. You can freely name the folder but it's recommended to name it with your wit ai bot name. You also can freely name your dataset files.

Folder structure:

|   app.py
|
└── your_bot
|   |   _config.yaml
|   |   dataset1.yaml
|   |   dataset2.yaml

0. Create your bot first through the wit.ai page, then get the BEARER code in the setting menu.

1. Open _config.yaml, write your BEARER code and version (the latest date) here

   Example:

   config:
     bearer: NKPU6CR7SCIHUZ5NR4762E2WXLQ6V3IP
     version: 15/01/2018
   

2. Define abbreviations of your intents and entities name in definition section in _config.yaml. This abbreviation will make it easier for you when labeling the entities of your data later. see description below for details of each key.

3. Create a Yaml dataset file (for example see sample_bot/all_data.yaml) which contains:

   • data
   • start (optional)
   • end (optional)

   see description below for details of each key

1. Define intents and entities name.

   Define abbreviations of your intents and entities name in definition section in _config.yaml. This abbreviation will make it easier for you when labeling the entities of your data later.

   WARNING! You can use any letter to make abbreviation. But to distinguish between intents and entities is required to start the abbreviation for the intents with the letter 'i'. So, the first letter of an abbreviation is 'i' if and only if it represents an intent.

   Example:

   definition:
     # INTENTS
     is: set
     iu: update
     ih: show
   
     # ENTITIES
     d: date
     m: month
     t1: hours
     t2: minutes
   

2. Tag the entities

   To use the abbreviations in your dataset, use backslash '\' before the part of the sentence you want to tag (LaTeX style). Use format:

   \abbrevations{tagged_text}`
   

   If you are using keyword entities in wit.ai you can assign a different value to an entity (which allows you to create synonims of keyword). To do that, add additional brackets right after tagged part of the sentence. Use format:

   \abbrevations{tagged_text}{assigned_value}
   

3. Tagging Example

   For example, if you have a sentence below and a list of abbreviations above.

   Set the date to 5 Feb, please
   

   to define the intent to "set", tag the whole sentence with set intent (is) abbreviation with formats like the example below:

   \is{Set the date to 5 Feb, please}
   

   Then, to define the date entities, tag the date part of the sentence with d:

   \is{Set the date to \d{5} Feb, please}
   

   In the example below we will tag "Feb" as month entities, but we assign it with value "february" (keyword entities):

   \is{Set the date to \d{5} \m{Feb}{february}, please}
   

   After you run it, the example above will train your bot with a data below:

   text : Set the date to 5 Feb, please
   intent : set
   date : 5
   month : february
   

Write your dataset in data section in your dataset file. There are two styles to write your dataset, numbered dataset or not numbered dataset. Please remember that you should only use one style.

1. Numbered dataset

   Numbered dataset needs another two keys: start and end, to define the start and the end of the part of data to be trained. It will be useful to avoid accidental redundancy of data train when training your bot. If you define the value of end with -1 then the data to be trained starts from the numbered data defined in start to the last. start and end are inclusive. See example in sample_bot/greetings_data.yaml.

   start: 1
   end: -1
   data:
     1: \ig{Good morning guys!}
     2: \ia{Who are you?!!}
     3: \is{Give me information about \n{Gilfoyle}}
     4: who is \n{Dinesh}?
     5: who are you \n{dinesh}{Dinesh}?
   

2. Unnumbered dataset

   The second type will always train the bot with all of your data (be careful with redundancy if you train your bot multiple times). See example in sample_bot/all_data.yaml.

   data:
     - \ig{Good afternoon guys!}
     - \ia{Who are you again?}
     - \is{Give me any information about \n{Gil}{Gilfoyle}}
     - \is{where is I can find King \n{Dinesh}}?
   

After define everything in _config.yaml and your dataset yaml file, simply train your bot by running the following command in the same directory with app.py and _config.yaml:

python app.py <bot_name> <dataset_name>

replace <bot_name> with your bot's folder name, and <dataset_name> with your specific dataset file name (with or without *.yaml). For example:

python app.py sample_bot greetings_data