Set y ranges in Parallel Coordinates plot

I want to compare many runs through a parallel coordinates plot and I need to evaluate them in terms of different metrics.
What I get by default when creating it manually is the following (where the various kinds of AUC scores is what I am interested in):

However, as you can see the AUCs columns have different yticks and range depending on the corresponding values, which makes it impossible to interpret them intuitively. Is it possible to set column ranges manually?

More context: I think this is related to another question on this blog. The reason why this would be convenient is when you want to compare different metrics or the same metric for different models.
In particular, I want to compare the performance of a baseline solution (Cumulative AUC (MELD) ) VS my model (Cumulative AUC). So intuitively I would see whether the lines go up passing from one column to the other to determin whether the second column shows better performance. Unfortunately, that is misleading if I the two columns have different ranges.

Hi @lclissa, thanks for your feedback! I also answeres the other question, so I am aware of this and currently it is not possible, so I will create a new feature request, thank you very much for providing the context. As I suggested in the other thread, a workaround could be to create two new runs logging the limits for each metric. Please let me know if I can help you in any other way!

1 Like

I can confirm that logging the endpoints works, thanks! :smiley:

Unfortunately, the workaround only partially solves the problem. In fact, this approach breaks when I need to perform dynamic filtering.

For example, I logged 0.5 and 1 as endpoints for all AUC metrics and it works for the “global” view.

However, for some reasons I then want to inspect the parallel coordinates plot of the best/worst runs. In this case, the filtered view would have unmatched lower/upper range endpoints respectively.

Of course, it would be impossible to manually log the endpoints for each possible filter, so I think the only way to handle this situation would be by setting a specified range for each column (unless one could log directly one line per run in the same parallel coordinates plot).

Ideally, it would be nice if the range could be set:

  • programmatically when logging + editable from the dashboard
  • optionally for groups of metrics
  • with a smart dynamic default (e.g. [min(group_metrics) - offset, max(group_metrics) + offset] )

I would be happy to contribute if needed, although I would need some guidance as I am not so expert of a programmer :slight_smile:
P.S. Can you link me the feature request so I can track it?

Hi @lclissa, thanks for explaining this! I’ll share this feedback with our Product Team. Regarding the feature request, unfortunately this is something internal, so I cannot share the link to it. We’ll notify you if this is implemented in the future.

1 Like

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