causalCannibalisation

Work in progress

Repository hosting the code for the IEEE paper Causal Quantification of Cannibalization during Promotional Sales in Grocery Retail

Datasets

• Corporacion Favorita Kaggle

• Breakfast at the Frat Dunnhumby

Libraries

• Papermill from nteract, pip install papermill
• CausalImpact from Dafiti pip install pycausalimpact

Analyse Dunnhumby data

From the csv data to the organised data (01.02.2021) Dunnhumby_arrange_store_sales.ipynb

To summarise the sales per store (prior to the analysis) Dunnhumby_Summarise_store_sales.ipynb

base NB to calculate the cannibalisation Dunnhumby_CausalImpact_Analysis_base.ipynb

Analyse Corporacion Favorita data

Use Papermill to create all the department NBs from a base one.

• base one: runner_papermill_causal_impact_covariates.ipynb It calculates all the potential cannibals and haloes
• autogenerated: papermill_unsupervised_{DEPARTMENT}_{STORE}.ipynb

To create the graphs showed in the paper

The chart showing the STL decomposition of the total sales generated with CFAV_store_sales_projection(paper).ipynb

To summarise the sales per store (prior to the analysis) CFAV_Summarise_store_sales.ipynb

To summarise all the results summarise_all_causal_results.ipynb For Dunnhumby data, use Dunnhumby_summarise_all_causal_results.ipynb

To run the surrogate model experiment, Surrogate_model_experiment_paper.ipynb

To explain how to select the promos CFAV_show_selection_for_CausalImpact.ipynb

To produce the cannibalisation episode plot CFAV_CausalImpact_Analysis_Dairy_one_case(paper).ipynb

To produce the cannibalisation episode using the Dunnhumby data, Dunnhumby_CausalImpact_Analysis_Paper_plot.ipynb

To produce the graph used in the paper CFAV-causal_impact_GROCERY_I_Pichincha_49_A_11(graph-paper).ipynb

Structure of the repo

The repo is structured as follows:

.
├── README.md     <- This file ;)
│
├── src
│   │
│   ├── notebooks   <- Collection of notebooks
│   ├── notebooks/preprocessing_envelope_for_seasonality.ipynb <- STL preprocessing
│   ├── notebooks/
│   ├── notebooks/

Installation as a Python wheel package

To generate the package from the source

python3 setup.py sdist bdist_wheel

The wheel can be directly imported in Databricks.