Loading…
Glossary term
Recording the settings and results of training runs so they can be compared and reproduced.
Training involves many runs with different settings. Experiment tracking records the parameters, code version, data and resulting metrics of each run, so results can be compared and reproduced. Dashboards show curves and let teams pick the best run.
Log a run with MLflow
import mlflow
with mlflow.start_run():
mlflow.log_param("learning_rate", 3e-4)
mlflow.log_metric("val_loss", 0.42)Without it, it is easy to lose track of what actually produced a good result.