M5 Forecasting - accuracy

This repository consists the solution for the M5 forecasting challenge. This competition was launched in March 2020. The competition aimed at predicting future sales at the product level, based on historical data.

Table of Contents

1. Setting up Python environment
2. Installing packages
3. Training the model with cross validation
4. Inference

1. Setting up python environment

It is recommended that you set up a virtual environment using the following steps: For MacOS users, you can run the following commands:

# Create a virtual env called order-allocation
pyenv virtualenv 3.7.9 m5_forecasting
# Activate the virtual env
pyenv shell m5_forecasting
# Update pip
pip install --upgrade pip

2. Installing packages

First clone the m5 repo. Install the packages in requirements.txt using

pip install -r requirements.txt

3. Training the model

If you haven't done so download the data from the kaggle competition website or using the Kaggle API and place it in the data folder. Change to the src folder as all the python scripts are there. To generate the necessary data frames for training and inference you need to run the feature_creation.py file first.

python feature_creation.py

To train the model with cross validation run the command:

python train_crossval.py

Since its training on a large dataset it will take a few hours and consume all the memory on your local machine. There is a saved model from the cross validation output in the model folder whihc you can use.

4. Generating forecasts

Run the main.py file to generate forecasts. The main.py uses the already trained model to do inference on a test set and outputs the rmsse for the test set as well as that of baseline model and a csv with the output forecast. You can also run the notebook, which does the same thing. The notebook has additional plots showing model performance and feature importance. To run the main.py file:

python main.py