README.md

Machine Learning end to end development

This repo contains a blueprint of the end to end process for the training, deployment and monitoring of machine learning models.

Scenario

For the blueprint we will use the Graduate Admission Dataset from Kaggle. We want to create some models which should be served in parallel. Also we want to be able to monitor the model performance and hot plug new models and retire old models.

Prerequisites

• Optional: Install PyCharm
• Install Anaconda
• Install Docker
• Install Docker-Compose (only on Linux)

Training

Build

cd training
conda env create -f environment.yml --prefix ./envs
conda activate ./envs
python setup.py [develop|install]

Run

cd training
conda activate ./envs
run_training

The new model will be stored at models/graduate-admissions/linear-regression-v1.pkl

Serving

Setup

Start Kafka cluster including Zookeeper, Kafka REST Proxy, Schema Registry and Admin Web UIs. See infrastructure/kafka/README.md for details.

Build

cd model-serving
conda env create -f environment.yml --prefix ./envs
conda activate ./envs
python setup.py [develop|install]
docker build -t model-server .

Run

conda activate ./envs
docker run model-server

Using Conda environment

Each subproject contains its own Conda environment file named conda.yml. You can apply the configuration using the following commands:

cd [subfolder]
conda env create -f environment.yml --prefix ./envs
conda activate ./[env-name]

You can find the name of the environment in the first line of the yaml file.

Adding new dependencies

Just add the dependency to the environment.yml manually. Afterwards you can import the new dependency using conda env update -f environment.yml --prefix ./envs