Continuous Intelligence Workshop

Running Workshop with dvc

1. Build the pipline
2. Push the results
3. Pull and reproduce

Build Pipeline with dvc

dvc run -d src/download_data.py -o data/raw/store47-2016.csv python src/download_data.py
dvc run -d data/raw/store47-2016.csv -d src/splitter.py -o data/splitter/train.csv -o data/splitter/validation.csv python src/splitter.py
dvc run -d data/splitter/train.csv -d data/splitter/validation.csv -d src/decision_tree.py -o data/decision_tree/model.pkl -M results/score.txt python src/decision_tree.py

Push the Results

First, push code changes to Github as usual, for instance:

git commit -am "Change model to be more awesome"
git push origin master

Next, push your dvc files to the cloud:

dvc push

That's it! Now anyone with access can fetch this repository and use dvc to replicate and build on your work.

Pull and reproduce changes

First, clone/pull this git repo.

git pull origin master --rebase

Next, pull from the cloud with dvc:

dvc pull

Finally, to reproduce the entire pipeline, simply run:

dvc repro model.pkl.dvc

Here, model.pkl.dvc is the last output in the dvc pipeline. Running it will reproduce all steps.

If you want to change the model, for example, edit the decision_tree.py file as you see fit. Then, you should be able to re-execute the model simply by re-running the pipeline using dvc repro model.pkl.dvc.

Local setup

Build locally and run

Once the model has been trained

docker build . -t ci-workshop

docker run -d -p 5005:5005 ci-workshop

You can view the app at http://localhost:5005

Note: try to assign 8G memory and 2CPU in Docker when running the docker build

Run with existing image

docker pull TBD

docker run -d -p 5005:5005 TBD