Sales Forecasting on Grocery Sales

This repository contains the codebase to reproduce the sales forecasting challenges on Corporación Favorita and Walmart.

Corporación Favorita Grocery Sales Forecasting

The challenge is presented in this Kaggle page. A description of the codebase with respect to analytics lifecycle below.

Required packages.

• python 3.X (Tested with 3.7)
• pandas
• numpy
• scipy
• scikit-learn
• lightgbm
• tqdm
• matplotlib
• squarify
• tensorflow 2.x

Codebase for analytics lifecycle

1. Obtaining environmental data.
   Weather data was collected from the World Weather Online API to enrich the Kaggle dataset. The scrpit to access the API can be found from favorita/Get_Temperature_Data.ipynb. You can download the acquired weather data from this Google drive link.

2. Data preparation.
   Basic pre-processing steps can be found from the first half of the Jupyter notebook at favorita/1_EDA_Cleaning.ipynb.

3. Data Exploration.
   Exploratory data analytics (EDA) is detailed in the second half of the same above notebook - favorita/1_EDA_Cleaning.ipynb.

4. Modeling prototypes.
   Prior to the model development, a prototyping was conducted for LGBM and DNN using Google Colab. Notebooks are available at favorita/2_Modeling_LGBM_Log_Scaled_Prototype.ipynb and favorita/3_Modeling_NN_Log_Scaled_Prototype.ipynb. Base code for LGBM and XGBoost are available at favorita/base_lgb_model.py and favorita/base_xgb_model.py.

5. Utility scripts.
   Script to load data: favorita/load_data.py
   Script to engineer features: favorita/feature_extractor.py
   Script to evaluate: favorita/evaluation.py
   !Important: Please create a config.py file in your environment indicating the root folder for the dataset.

6. Hyper-parameter search.
   Random Search and Grid Search scripts for LGBM can be found at favorita/base_lgb_model_random_search.py.

7. Predictive models (general model for all stores).
   Script for LGBM: favorita/model_lgbm.py
   Script for DNN: favorita/model_nn.py

8. Predictive models (per store model).
   Script for LGBM: favorita/model_lgbm_per_store.py
   Script for DNN: favorita/model_nn_per_store.py

9. Ensemble.
   Prototype ensemble is avaialble here: favorita/4_Modeling_LGBM_Ensemble.ipynb