Histogram over time

Hi all,

I would like to have a plot of multiple histograms, that I log over time, like in the figure attached. They are the activations of units in a layer, thus ‘wandb.watch()’ doesn’t work, as it is only for gradients and parameters, right?

I’m using the standard histogram code, and it shows the latest version of the histogram, not all of them over time.
“”“”
def hook(model, input, output):
data = [[x.item()] for x in output.flatten()]
table = wandb.Table(data=data, columns=[“activations”])
wandb.log({f’layer_hist’: wandb.plot.histogram(table, “activations”,
title=f"activations")})
“”"

Is there a way to get the histogram over time plot?

Thanks, Carlos

Hi @cstein06,
If you go to edit the chart and change from the summary table to the history table. Then you can edit the chart specs and add a step slider to visualize this over time. Let me know if this helps.


1G19Vs7tcBzpH15GSvJgIemBYmt.png)

Hi @cstein06, I’ll close this out since I believe this should solve your issue but feel free to ping me here if you have any other questions or this doesn’t resolve this for you.

Thank you,
Nate

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.