Deployment for the Accent Training project

This project is used to deploy a telgeram bot and an accent prediction algorithm. It is additionally used to learn and train the model.

There are two optional compose files

1. keras_training
   1. Train and learn the CNN on sound files
2. prediciton_bot
   1. Serves the model as a microservice and starts a Telegram chatbot that communicates with it.

The project is seperated into two docker-compose files for convinice which share a volume for the saved models

Development Installation

• Clone the following repositories in the same directory:
  • accent-bot
  • keras-accent-trainer
  • accent-training-deployment

Deploying keras_training

To load the artifacts and create the csv file:

docker-compose -f "keras_training/docker-compose.yml" up --build

To run the prediction

docker-compose -f "keras_training/docker-compose.predict.yml" up

Choosing a Prediction model

The compose file for production chooses the ID served model. When running through compose the saved models are located in /share/saved_models It follows the Comet-ML experiment IDs folder scheme

├── saved_models              - parent directory shared in a volume
    ├── <experiment name>     - this folder contains all experiments in a certain type
        ├── <experiment id>   - multiple folder for every experiement.
            |── model.h5      - save Keras model weights

Deploying prediction_bot

1. Update BOT_ID in docker-compose.production.yml with your Telegram Bot ID
2. Update MODEL_TYPE and MODEL_NUM in docker-compose.production.yml with the models you want to serve The model depend on the model you trained in the previous stage
3. Stop tracking docker-compose.production.yml:

   git update-index --assume-unchanged docker-compose.production.yml
   

4. To continue tracking the file docker-compose.production.yml:

   git update-index --no-assume-unchanged docker-compose.production.yml
   

5. run:

   docker-compose -f "prediction_bot/docker-compose.yml" -f "prediction_bot/docker-compose.production.yml" up -d --build