I have my MLflow runs set up to log train/val loss regularly, like this:
mlflow.log_metric("train/loss", train_loss, step=iter_num)
In the web view I can click “model metrics” and see a plot of loss by step. But when I try to compare the loss curve between two runs by selecting them and clicking “compare”, I can’t find any visualization that shows more than a single value for the loss (the loss from the last step logged, I think).
Is there a way to see the same kind of plot I get in “model metrics” but comparing two runs? Am I using the API wrong?
-
use
mlflow.create_experiment()to make a new experiment -
for two different runs, run using
mlflow.start_run() -
log loss each step/epoch by calling
mlflow.log_metric(..., step=X) -
in web UI, check the two runs and click compare
-> all but the loss value form the final step seems to have disappeared
haltil is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.