Comparing metrics and plots in DVC

CI/CD for Machine Learning

Ravi Bhadauria

Machine Learning Engineer

Configuring DVC YAML file

Configure DVC YAML file to track metrics across experiments
Change from outs

stages:
  preprocess:
    ...
  train:
    ...
    outs:
    - metrics.json
    - confusion_matrix.png

To metrics

stages:
  preprocess:
    ...
  train:
    ...
    outs:
    - confusion_matrix.png

    metrics:
      - metrics.json:
          cache: false

Querying and comparing DVC metrics

-> dvc metrics show

Path          accuracy    f1_score    precision    recall             
metrics.json  0.947       0.8656      0.988        0.7702

Change a hyperparameter and rerun dvc repro

-> dvc metrics diff

Path          Metric     HEAD    workspace    Change                  
metrics.json  accuracy   0.947   0.9995       0.0525
metrics.json  f1_score   0.8656  0.9989       0.1333
metrics.json  precision  0.988   0.9993       0.0113
metrics.json  recall     0.7702  0.9986       0.2284

¹ https://dvc.org/doc/command-reference/metrics

Setting up DVC Github Action

Add setup-dvc GitHub Action
Replace running Python scripts with DVC pipeline

steps:
  ...
  - name: Setup DVC
    uses: iterative/setup-dvc@v1

  - name: Run DVC pipeline
    run: dvc repro

Setting up DVC Github Action

- name: Write CML report
  env:
    REPO_TOKEN: ${{ secrets.GITHUB_TOKEN }}
  run: |
    # Print metrics of current branch
    dvc metrics show --md >> report.md

    # Compare metrics with main branch
    git fetch --prune
    dvc metrics diff --md main >> report.md

    # Create CML report
    cml comment create report.md

Pipeline in action

Screenshot of a pull request comment showing metrics diff against main branch

Plot types in DVC

scatter - scatter plot
linear - interactive linear plot
simple - non-interactive customizable linear plot
smooth - linear plot with smoothing
confusion - confusion matrix
confusion_normalized - confusion matrix with values normalized to <0, 1> range
bar_horizontal - horizontal bar plot
bar_horizontal_sorted - horizontal bar plot sorted by bar size

¹ https://dvc.org/doc/user-guide/experiment-management/visualizing-plots#plot-templates-data-series-only

Configuring DVC YAML for plots

stages:
  train:
    ...
    plots:
    - predictions.csv: # Name of file containing predictions
        template: confusion # Style of plot

        x: predicted_label # X-axis column name in csv file 
        y: true_label # Y-axis column name in csv file
        x_label: 'Predicted label'
        y_label: 'True label'
        title: Confusion matrix

        cache: false # Save in Git

Plotting Confusion Matrix

-> dvc plots show predictions.csv
file:///path/to/index.html

Confusion matrix display plot generated by DVC

Comparing Confusion Matrix

-> dvc plots diff --target predictions.csv main
file:///path/to/index.html

Confusion matrix diff plot generated by DVC

Comparing ROC Curves

# Changes in Python
y_proba = model.predict_proba(X_test)
fpr, tpr, _ = roc_curve(y_test, 
                        y_proba[:, 1])

# Changes in dvc.yaml
plots:
- roc_curve.csv:
    template: simple
    x: fpr
    y: tpr
    x_label: 'False Positive Rate'
    y_label: 'True Positive Rate'
    title: ROC curve
    cache: false

ROC curve diff plot generated by DVC

Let's practice!

CI/CD for Machine Learning