plots.md

Plots Dashboard

Select (click the circle beside the experiment name) up to 7 experiments in the table view to visualize in the Plots Dashboard.

Use DVC: Show Plots from the command palette to open it or open it using the table's row context menu.

💡 To add DVC plots to the project, start writing data series into JSON, YAML, CSV, or TSV files; or save your own plot images (.png, etc.). Use DVC: Add Plot to define plots in a dvc.yaml file. If you're using Python, the DVCLive helper library can save plots data for you!

points = metrics.precision_recall_curve(labels, predictions)
with open("plots.json", "w") as fd:
     json.dump({"prc": [
          {"precision": p, "recall": r, "threshold": t}
               for p, r, t in points
     ]})

from matplotlib import pyplot as plt
fig, axes = plt.subplots(dpi=100)
...
fig.savefig("importance.png")

from dvclive import Live
live = Live("evaluation")
live.log_plot("roc", labels, predictions)

These are the types of plots that can be displayed (for the selected experiments):

Data Series are JSON, YAML, CSV, or TSV files visualized using plot templates, which may be predefined (e.g. confusion matrix, linear) or custom (Vega-lite files)

Images (e.g. .jpg or .svg files) can be visualized as well. They will be rendered side by side for the selected experiments.

Custom plots compare a chosen metric and param across experiments.

The Plots Dashboard can be configured and accessed from the Plots side panel in the DVC View.

This is equivalent to the dvc plots show and dvc plots diff commands.