Starter kit for projects from Approaching Almost Any Machine Learning problem book by Abhishek Thakur
This starter is very much work in progress. I am updating key modules as I am making my way through the book.
- install miniconda
conda create -n NAME_FOR_YOUR_CONDA_ENV python=3.7.6
(if you are on M1/M2, you might need to run this beforeconda config --env --set subdir osx-64
)conda activate NAME_FOR_YOUR_CONDA_ENV
conda env create -f environment.yml
- OR
conda env create -f environment_osx.yml
on Mac conda activate ml
- download
mnist_train.csv
from Abhishek's Kaggle, and save it toinput
directory
This template gets you started with a basic skeleton for your ML app.
cd src && python create_folds.py --folds 10
(default is 5 folds)
model_dispatcher.py
contains a dictionary with models available in your application
To train your model, use
python train.py --fold 0 --model desision_tree_entropy