Tracking metrics should support batch sizes > 1 #272
Labels
comp: metrics
Metrics to evaluate what we're doing
effort: high
objective: speed
Affecting training speed
prio: low
The tracking metrics (evaluating the quality of the clustered latent space) currently only make sense when they are fed single graphs (rather than batches of multiple graphs). This slows down the test step.
Low prio because we're memory constrained anyway.
The text was updated successfully, but these errors were encountered: