From POC to Production

Introduction

This repository stores my first work on building a ML model and deploying into a production environment.

The goal consists on building a model that predict a stackoverflow tag based on an input text. The model is divided in 3 parts:

• Preprocessing : Package with useful functions used by the model for both training and prediction. We herited the Sequence Class from Keras.
• Train : Train the deep learning model.
• Predict Load the trained model then predict the input text.

After building the model, we create a flask application so that we can send the input text via a HTTP request.

Then we want to dockerize our application so that it can be deployed on any machine.

Getting started

First you need to clone the repository:

git clone https://github.com/GabrielTruong/poc2prod_capstone.git

Then you need to create a python conda environment:

conda create -n ENVNAME --file environment.yml

Deploy the app locally without docker:

python predict/predict/app.py

The application will run at the following adress: http://127.0.0.1:5000/

Use the app

1. Using the form

Use the link http://127.0.0.1:5000/ and use the form to predict the input text.

2. Using a curl request

The route is under http://localhost/predict. It is a POST route that needs a JSON body with 2 specific arguments:

• texts : a list of string. Each string value correspond to the text you want to predict.
• top_k : The limit number of the prediction you want to return

POST

curl -X POST -H "Content-Type: application/json" -d '{"text": [TEXT1,TEXT2], "top_k":NUMBER_OF_CLASS}' http://localhost:5000/predict

You can use a simple HTTP GET request by sending the input text via the parameter string text as shown below: GET

curl http://127.0.0.1:5000/get_prediction?text=<INPUT TEXT>

Useful command

1. Test on predict

python -m unittest predict.tests.test_predict

2. Test on train

python -m unittest train.tests.test_model_train

3. Test on preprocessing

python -m unittest preprocessing.tests.test_utils

Problem with module error

add python files to python path

export PYTHONPATH=$PYTHONPATH:/home/dev/python-files

export PYTHONPATH=$PYTHONPATH:/home/lapbeer/Documents/epf/5A/poc2prod/poc2prod_capstone

Train model

python train/train/run.py "train/data/training-data/stackoverflow_posts.csv" train/conf/train-conf.yml "train/data/artefacts/train/"

Activate conda environment

conda activate ENVNAME

CHECK SETUPTOOLS

RESTE à FAIRE:

• Revoir les path pour l'artefact
• Revoir le fichier predict car bancal
• Faire fonctionner l'app flask avec curl et UI
• Dockeriser l'app
• Airflow