Hello, I have been unable to make a bar chart showing the differences between a set of fixed “True” values and the final/summary output of their predicted counterparts within a sweep. Can this be done in wandb and if so, how?
My approach was to use ${summary:Train_loss} - ${Train_exact} in the expression panel of a bar chart which had Train_loss as the x data input. I have also tried simple commands such as subtracting scalar values from the summarized loss but those have also failed.
I recently received an email notification from the helpdesk but there is no response content, just a copy of my original post. I am not sure if this is an error or an unknown quirk of how wandb.ai works. Please advise.
Hey @cr_97, I am digging into this issue further, and will update you as progress arises. For now I recommend the following workaround:
Exporting Train_loss and Train_exact as CSV files, creating the bar chart in your Python script, then uploading that to wandb (or alternatively uploading a table of the differences to wandb and generating the bar chart in our UI)