Inspecting Training Results#

The return value of your call is a Result object.

The Result object contains, among other information:

  • The last reported metrics (e.g. the loss)

  • The last reported checkpoint (to load the model)

  • Error messages, if any errors occurred

Viewing metrics#

You can retrieve metrics reported to Ray Train from the Result object.

Common metrics include the training or validation loss, or prediction accuracies.

The metrics retrieved from the Result object correspond to those you passed to as an argument in your training function.

Last reported metrics#

Use Result.metrics to retrieve the latest reported metrics.

result =

print("Observed metrics:", result.metrics)

Dataframe of all reported metrics#

Use Result.metrics_dataframe to retrieve a pandas DataFrame of all reported metrics.

df = result.metrics_dataframe
print("Minimum loss", min(df["loss"]))

Retrieving checkpoints#

You can retrieve checkpoints reported to Ray Train from the Result object.

Checkpoints contain all the information that is needed to restore the training state. This usually includes the trained model.

You can use checkpoints for common downstream tasks such as offline batch inference with Ray Data, or online model serving with Ray Serve.

The checkpoints retrieved from the Result object correspond to those you passed to as an argument in your training function.

Last saved checkpoint#

Use Result.checkpoint to retrieve the last checkpoint.

print("Last checkpoint:", result.checkpoint)

with result.checkpoint.as_directory() as tmpdir:
    # Load model from directory

Other checkpoints#

Sometimes you want to access an earlier checkpoint. For instance, if your loss increased after more training due to overfitting, you may want to retrieve the checkpoint with the lowest loss.

You can retrieve a list of all available checkpoints and their metrics with Result.best_checkpoints

# Print available checkpoints
for checkpoint, metrics in result.best_checkpoints:
    print("Loss", metrics["loss"], "checkpoint", checkpoint)

# Get checkpoint with minimal loss
best_checkpoint = min(
    result.best_checkpoints, key=lambda checkpoint: checkpoint[1]["loss"]

with best_checkpoint.as_directory() as tmpdir:
    # Load model from directory

See also

See Saving and Loading Checkpoints for more information on checkpointing.

Accessing storage location#

If you need to retrieve the results later, you can get the storage location of the training run with Result.path.

This path will correspond to the storage_path you configured in the RunConfig. It will be a (nested) subdirectory within that path, usually of the form TrainerName_date-string/TrainerName_id_00000_0_....

The result also contains a pyarrow.fs.FileSystem that can be used to access the storage location, which is useful if the path is on cloud storage.

result_path: str = result.path
result_filesystem: pyarrow.fs.FileSystem = result.filesystem

print("Results location (fs, path) = ({result_filesystem}, {result_path})")

You can restore a result with Result.from_path:

from ray.train import Result

restored_result = Result.from_path(result_path)
print("Restored loss", result.metrics["loss"])

Viewing Errors#

If an error occurred during training, Result.error will be set and contain the exception that was raised.

if result.error:
    assert isinstance(result.error, Exception)

    print("Got exception:", result.error)

Finding results on persistent storage#

All training results, including reported metrics, checkpoints, and error files, are stored on the configured persistent storage.

See our persistent storage guide to configure this location for your training run.