Trial metrics

With the dvc metrics command, DVC is also a framework for recording and comparing the performance of experiments. calculates the AUC (A rea U nder the C urve). It uses the test data set, reads the features from the file features/test.pkl and creates the metrics file auc.metric. It can be identified as a DVC metric with the -M option of dvc run, in our example with:

$ dvc run -n evaluate -d src/ -d model.pkl -d data/features \
    -M auc.json python src/ model.pkl data/features auc.json
  cmd: python src/ model.pkl data/features auc.json
  - data/features
  - model.pkl
  - src/
  - auc.json:
      cache: false

With dvc metrics show experiments can be compared then through various branches and tags:

$ dvc metrics show
        auc.json: 0.514172

Now to complete our first version of the DVC pipeline, let’s add the files and a tag to the Git repository:

$ git add dvc.yaml dvc.lock auc.json
$ git commit -m 'Add stage ‹evaluate›'
$ git tag -a 0.1.0 -m "Initial pipeline version 0.1.0"