DiscriminationThreshold seems to compute the minimum number of thresholds across trials, not max

Presumably the following code should find the max number of thresholds found across all the trials so as not to ignore the values in the densest threshold list across all trials:

# Compute maximum number of uniform thresholds across all trials
n_thresholds = np.array([len(t['thresholds']) for t in trials]).min()
self.thresholds_ = np.linspace(0.0, 1.0, num=n_thresholds)

But this finds the minimum. So we may be losing information by bisecting with the sparsest number of thresholds rather than the densest.

[bbengfort]

@aronszanto glad to hear that you’re using Yellowbrick in your work! Do you mind if I ask where you work? (We’re just interested in how our users are using the library). Feel free to send me a DM on twitter (@bbengfort) if you want to chat about it!

If we can be of any help with your company lawyers, please let us know! Yellowbrick is licensed under a permissive Apache 2 open source license which means you can absolutely use it for commercial use and make modifications to the software. All we ask is that you include our copyright, license, and any notices along with your fork of the code. If you make any major changes to the code, we’d also ask that we were notified.

Hopefully your managers will agree with me that contributing to open source projects like YB enhances the value of your work and will let you open a PR with the fix you described. If they don’t, we understand - we are all volunteers doing this in our spare time and also have full time jobs that let us contribute more or less to other open source projects!

[aronszanto]

Hey all, thanks for bumping. I discovered this and then wrote a fix for my company, which means that whether I can open source it is up to the lawyers. Working on this 😃