SpawN ML Bot Backend V 1.0.0

The high scalable intent classification engine for chatbot. The model built is based on Bag of words approach. This is best suited for in domain classification task.

SpawN ML now support loading multiple models for inference for text classification.

Note:- The application can be run using Python3 only

• This repo contains the text classification back end written in flask
• This uses Tensorflows TFLearn library for building the neural network model
• The flask app is run using the wsgi wrapper server using Tornado for better concurrency
• This repo can be used for making high scalable backend for building the chatbots

1. Project Structure

The project has following structure:-

• models/ - This directory contains the models built by TFLearn. This directory is used for saving and loding the models. The directory also has a pickle.
• training_data - This folder contains the data file for training the neural network. Also the tflearn logs are stored inside this folder.
• SpawnMLBackend.py - Flask webservice python file. The file has also decorator written for authentication of the webservice. The authentication used is Basic Authentication. This can be changed as per your requirement.
• train_network.py - All the training logic,loading models at startup and classification logic is written in this file.
• tensorflow_async_server.py - Flask server is not suitable for deploying in production. The wsgi wrapper is written using Tornado webserver for high scalability.
• waitress_server.py - Waitress is meant to be a production-quality pure-Python WSGI server with very acceptable performance. You can either server the model using tensorflow_async_server or waitress_server

Note: Please change the directory inside train_network.py and keras_train.py to your specific model directory. In future release, this will be fixed. Also note for training and inference it currently uses keras in SpawnMLBackend.py. You can switch into tflearn by importing the train_network in flask app.

In my load test, I found Tornado webserver to be more stable than waitress. Waitress gave me higher throughput than Tornado webserver but Tornado webserver gave me better stability with zero failed request.

2. Running application

To run this application , install the depencencies as:

• pip install tensorflow
• pip install tflearn
• pip install flask
• pip install tornado
• pip install nltk - you will also need to download 'punkt' of nltk.
• pip install pathlib

After installing all the dependencies, run the app as:

• python3 tensorflow_async_server.py
• For background running task- nohup python3 tensorflow_async_server.py &

Note: The default authentication for testing is username=username, pass- password. You can change the authentication as per your requirement.

• Basic dXNlcm5hbWU6cGFzc3dvcmQ=

3. Links

• TFLearn
• Tensorflow
• Keras
• Tornado
• Waitress

Future Release: Docker support, Dynamic model building, Web UI for building the models.