RASA NLU Setup

RASA NLU Setup

Rasa NLU is a powerful open-source natural language understanding tool that enables chatbots and virtual assistants to understand and interpret user messages. In this guide, we’ll walk you through the setup of Rasa NLU, including installing dependencies, training models, and integrating with various backends. Additionally, we’ll explore date parsing using Duckling-rest.

install pip if not installed

sudo apt-get update
sudo apt-get install python-pip

clone the repo and install from setup.py

git clone https://github.com/golastmile/rasa_nlu.git
cd rasa_nlu
sudo python setup.py install

Download backends options

MITIE

Download and save total_word_feature_extractor.dat file. Copy total_word_feature_extractor.dat file from here :https://github.com/mit-nlp/MITIE/releases/download/v0.4/MITIE-models-v0.2.tar.bz2.

mkdir mitie
mv /path/to/total_word_feature_extractor.dat rasa_nlu/mitie/
pip install git+https://github.com/mit-nlp/MITIE.git

spaCy and sklearn

pip install -U spacy
python -m spacy.en.download all
pip install numpy scipy scikit-learn

Training

Saving models

mkdir models
cd models
mkdir wit
mkdir rasa
mkdir data
cd data
mkdir wit
mkdir rasa

Training data

You may use online rasa nlu trainer to create your training data here https://golastmile.github.io/rasa-nlu-trainer/ After adding your data, download the training data from the trainer. save the file in data/rasa/ folder.

rasa

Create a config file called config_rasa.json with below contents:

{
  "backend": "mitie_sklearn",
  "mitie_file": "/home/<user>/rasa/rasa_nlu/mitie/total_word_feature_extractor.dat",
  "path" : "/home/<user>/rasa_nlu/models/rasa/",
  "data" : "/home<user>/rasa_nlu/data/rasa/rasa.json"
}

wit

Extract the data from your wit project and copy the expressions.json in data/wit folder. Create a config file called config_wit.json with below contents:

{
  "backend": "mitie_sklearn",
  "mitie_file": "/home/<user>/rasa_nlu/mitie/total_word_feature_extractor.dat",
  "path" : "/home/<user>/rasa_nlu/models/wit/",
  "data" : "/home<user>/rasa_nlu/data/wit/expressions.json"
}

Training

Firsts training rasa use config_rasa.json, then run the same command with config_wit.json

python -m rasa_nlu.train -c config_<rasa/wit>.json

After successful training, your models will be created in the models folder for both configurations.

Running Server

Create a config file config_server.json with following contents:

{
  "mitie_file": "/home<user>/rasa_nlu/mitie/total_word_feature_extractor.dat",
  "server_model_dirs": {
    "wit" : "/home<user>/rasa_nlu/models/wit/<model_folder>/"
  },
  "token": <token>,
  "response_log":"/home<user>/rasa_nlu/logs/",
  "emulate" : "wit" 
}

Start server

python -m rasa_nlu.server -c config_server.json

Making requests

You can make GET or POST requests to the server. Here’s an example of a GET request:

GET "http://localhost:5000/parse?q=<text>&model=<model>&token=<token>"

Date parsing with Duckling-rest

clone the repo

git clone https://github.com/ShubhankarS/duckling-rest.git

Install java version 6 or later if not installed. Copy lein script from here https://raw.githubusercontent.com/technomancy/leiningen/stable/bin/lein

Follow below steps to execute the script:

cd /bin
sudo vi lein
Copy the contents of the script into lein
sudo chmod 755 lein
lein

duckling-rest set up

cd duckling-rest/duckling
lein jar
lein install
cd ..
lein deps
lein run

Making requests

You can make requests to Duckling-rest for time parsing:

GET "http://localhost:5000/parse/time/:text"

With this guide, you’ve successfully set up Rasa NLU with different backends and integrated date parsing using Duckling-rest. You’re now ready to build powerful natural language understanding capabilities into your applications.