Setup

Read the complete tutorial here

git clone git@github.com:curiousily/Reproducible-ML-with-DVC.git

pipenv install --dev

git checkout pre-dvc

DVC

Initialize DVC

dvc init

and add remote storage (local in this case)

dvc remote add -d localremote /tmp/dvc-storage

disable analytics (optional)

dvc config core.analytics false

Experiment with Linear Regression

Build Dataset

dvc run -f assets/data.dvc \
    -d studentpredictor/create_dataset.py \
    -o assets/data \
    python studentpredictor/create_dataset.py

Create features

dvc run -f assets/features.dvc \
    -d studentpredictor/create_features.py \
    -d assets/data \
    -o assets/features \
    python studentpredictor/create_features.py

Train model

dvc run -f assets/models.dvc \
    -d studentpredictor/train_model.py \
    -d assets/features \
    -o assets/models \
    python studentpredictor/train_model.py

Evaluate the model and save metrics (RMSE and r^2)

dvc run -f assets/evaluate.dvc \
    -d studentpredictor/evaluate_model.py \
    -d assets/features \
    -d assets/models \
    -M assets/metrics.json \
    python studentpredictor/evaluate_model.py

Check the metrics for your current model:

dvc metrics show -T

Experiment with Random Forest

Checkout the Random Forest experiment:

git checkout rf-experiment

Reproduce everything with the RF model

dvc repro assets/evaluate.dvc

Check the metrics for the Random Forest model compared to the Linear Regression:

dvc metrics show -T

Read the complete tutorial here

License

MIT