Machine Learning Exercise

This repository contains an attempt of a well-organized machine learning (MLE) project, focusing on achieving compatibility with different machine environments.

Overview of Functionality

1. Data Generation: Generate the data that the model will be trained and tested on. For simplicity's sake, the Iris dataset will be used.

2. Model Training: The model is trained using the train.py script, which uses the iris_train.csv dataset.

3. Inference on Unseen Data: The model is used to predict outcomes using unseen data from the iris_test.csv file. It ouputs the results to a .csv file.

Usage

Docker

1. Clone the repository

   git clone https://github.com/tunjan/mle-homework.git

2. (optional) Open VSCode, go to the File menu, and select Add Folder to Workspace.

3. (optional) Navigate to the cloned repository's directory and add it.

4. Set the CONF_PATH environment variable to "settings.json".

   export CONF_PATH="settings.json"

5. Run the bash script from the parent folder (use sudo if necessary).

   bash script.sh

6. The generated model should be in the /models folder and the infered results on unseen data in the /results folder.

NOTE: The Shell script was tested both on Unix and Windows operating systems. The only dependencies are Git, Python and Docker. If you're using Microsoft Windows make sure to open the terminal on the folder with Git and Bash.

Locally

For local execution, run these scripts in sequence: data_process/data_generation.py, training/train.py, inference/run.py. The results should be similar to the Docker approach. However, using Docker is advised for consistent reproducibility across different operating systems. The trained model should be in the /models directory, and predictions on new data in the /results folder.

Considerations

It's worth noting that installing TensorFlow within the Docker container is a painful requirement, as it is a substantially big package (475 MB). Patience or a high download bandwidth may be required.

Dependencies

• Docker
• Python (pip)
• Bash

Attributions

Thanks to the original developer of the code. Any remaining mistakes, as is often inevitable, are my sole responsibility.