Closed crazyboy9103 closed 10 months ago
@crazyboy9103 Wouldn't it be easier to do this?
# Log the metric
self.log("my_metric", value)
# Also log min and max
self.log("my_metric_min", value, reduce_fx="min")
self.log("my_metric_max", value, reduce_fx="max")
I simply wasn't aware of this functionality, as it wasn't well explained in the doc. It is definitely easier, thank you for pointing out the detail in the doc.
Description & Motivation
It is sometimes important to monitor minimum/maximum values for results that is logged via self.log, e.g. losses, metrics. Although ModelCheckpoint saves checkpoints using "monitor", it seems convenient to log the min/max values and see how they change over time in Tensorboard/Wandb and others.
Pitch
No-brainer version i'm using currently
Alternatives
No response
Additional context
No response
cc @borda @awaelchli