Repository containing the code used for my presentation at Data Science Summit 2024, ML Edition
A demo project showcasing experiment tracking capabilities of DVC. The example is built on the case of classifying Pokémon as legendary.
- Generation is based on Pokédex number. That does not reflect regional forms (Galarian, Alolan, etc.), Mega Evolutions, etc.
- Until generation 8 (incl.), the legendary classification consisted of the following: Sub-Legendary Pokémon | Legendary Pokémon | Mythical Pokémon. We treat them all as legendary for the sake of this exercise.
- From generation 9 onwards, the legendary classification is slightly different, that is, there are groups called: Sub-Legendary Pokémon | Ultra Beasts | Paradox Pokémon | Restricted Legendary Pokémon | Mythical Pokémon. For example, Ultra Beasts are now a separate class and are not considered Legendary.
- prepare the raw dataset using the
src/getting_data/get_pokedex.py - execute the pipeline (or its steps) as described in
dvc.yaml
Docs:
My articles on setting up experimentation with DVC:
- https://towardsdatascience.com/turn-vs-code-into-a-one-stop-shop-for-ml-experiments-49c97c47db27
- https://towardsdatascience.com/enhance-your-ml-experimentation-workflow-with-real-time-plots-434106b1a1c2
- https://towardsdatascience.com/the-minimalists-guide-to-experiment-tracking-with-dvc-f07e4636bdbb
- https://towardsdatascience.com/keep-track-of-your-backtests-with-dvcs-experiment-tracking-38977cbba4a9
- https://towardsdatascience.com/experiment-tracking-hyperparameter-tuning-organize-your-trials-with-dvc-d17f47f38754
How DVC uses git references for experiment tracking:
Misc:
- https://www.kaggle.com/datasets/mariotormo/complete-pokemon-dataset-updated-090420/?select=pokedex_%28Update_04.21%29.csv
- https://pokemondb.net/pokedex/all
- https://www.serebii.net/pokemon/legendary.shtml
- https://www.serebii.net/pokemon/specialpokemon.shtml
Icons: