UXML for e-commerce

E-commerce recommender based on user events

How to use this repo?

Start with eda-uxml.ipynb for exploratory data analysis
Prepare the data for machine learning with data data-preparation-uxml.ipynb (Optionally test the resulting .npz with sparse-matrix-tester.ipynb)
Run one of the training algorithms, such as train-uxml-basic-matrix-factorization.ipynb (more to come in the future)
Test the performance of the training alogrithm against a test set (different from train and validation data) with test-uxml.ipynb or quick-test-uxml.ipynb
Put the results in practice. Two use case examples are provided use-case-examples.ipynb (item recommender for users and minimalistic stock need prediction to help with e-commerce logistics)

The Optuna based implementation of HPO: train-uxml-adam-optuna.ipynb
Applying the hyper-parameters found by Optuna without the need for Optuna framework: train-uxml-adam-from-best-trial.ipynb

The eda-fe-5months.ipynb represents scaling the orginal dataset for 5 months (instead of the original 2)
data-cleaning-rees46-purchase.ipynb introduces a new dataset with 123 event types (purchases from 123 different categories) collected between 2020-01-05 and 2020-11-21 by the REES46 Marketing Platform.
eda-fe-rees46-purchase.ipynb represents the generalization effort with an arbitrary number of event types from the cleaned up dataset

The data-preparation-test.ipynb can be used to test the efficiency of data preparation, as in comparing the prepared data to the ground truth. This is not needed for the user-behaviour prediction process.
The quick-test-uxml.ipynb relies on sparse matrix operations to do only MSE, RMSE, MAE, R-squared, and explained variance, for this reason it runs in 1.1s, compared to 175.89s of the full test which has multiple approaches, and many more metrics.

Name		Name	Last commit message	Last commit date
Latest commit History 89 Commits
data		data
utilities		utilities
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
constants.py		constants.py
data-cleaning-rees46-purchase.ipynb		data-cleaning-rees46-purchase.ipynb
data-preparation-test.ipynb		data-preparation-test.ipynb
data-preparation-uxml-time.ipynb		data-preparation-uxml-time.ipynb
data-preparation-uxml.ipynb		data-preparation-uxml.ipynb
eda-fe-5months.ipynb		eda-fe-5months.ipynb
eda-fe-rees46-purchase.ipynb		eda-fe-rees46-purchase.ipynb
eda-uxml.ipynb		eda-uxml.ipynb
quick-test-uxml.ipynb		quick-test-uxml.ipynb
requirements.txt		requirements.txt
sparse-matrix-tester.ipynb		sparse-matrix-tester.ipynb
speed-comparison.xlsx		speed-comparison.xlsx
test-uxml-best-trial.ipynb		test-uxml-best-trial.ipynb
test-uxml.ipynb		test-uxml.ipynb
train-uxml-adam-from-best-trial.ipynb		train-uxml-adam-from-best-trial.ipynb
train-uxml-adam-optuna.ipynb		train-uxml-adam-optuna.ipynb
train-uxml-adam.ipynb		train-uxml-adam.ipynb
train-uxml-basic-matrix-factorization.ipynb		train-uxml-basic-matrix-factorization.ipynb
use-case-examples.ipynb		use-case-examples.ipynb