Skip to content

Tensorboard

Sample Factory uses Tensorboard summaries. Run Tensorboard to monitor any running or finished experiments:

tensorboard --logdir=<your_train_dir> --port=6006

Monitoring multiple experiments

Additionally, we provide a helper script that has nice command line interface to monitor select experiment folders using wildcards:

python -m sample_factory.utils.tb --dir=./train_dir '*name_mask*' '*another*mask*'

Here --dir parameter is the root directory with experiments, and the script will recursively search for experiment folders that match the masks.

Monitoring experiments started by the Launcher

Launcher API is a convenient way to start multiple experiments in parallel. Such groups of experiments can be monitored with a single Tensorboard command, just specify --logdir pointing to the root directory with experiments.