This repository contains my experiments on the Rossmann Store Sales dataset using the fastai library.
The goal is to investigate if there are any general techniques (e.g. preprocessing steps, feature engineering, etc) that can help deep learning models on tabular data converge faster and/or with higher accuracy.
-
Install Docker
-
Navigate to this project's root directory
-
Build and run this project's Dockerfile by executing:
docker build -t fastai . && \ docker run --runtime=nvidia \ --name fastai \ -p 8888:8888 \ --dns 8.8.8.8 \ fastai jupyter notebook --ip 0.0.0.0 --allow-root
This will start a jupyter server on port 8888.
-
Open a web browser and navigate to the URL displayed in the previous step.
-
Navigate to the notebooks folder. This folder contains:
- 01_rossmann_data_clean: Download the training and test set
- 02_EDA: Some exploratory data analysis on the dataset
- 03_modelling: Models/techniques I've tried so far
Feel free to raise a pull request for any questions, comments, feedback, etc.