Iain Carmichael

Ok I upgraded numpy to version 1.22.3 and it now works. It might be useful to add a warning or comment around that import in `histomicstk/segmentation/label/trace_object_boundaries.py` so people know to...

Will do. May take sometime.

Absolutely! I'm going to submit a PR that adds the new AJIVE code as soon as the new MCCA module (#249) is added.

Great to hear! I agree reweighted L1 is a good name i.e. tells you exactly what the algorithm does. That being said the stats literature tends to use LLA (e.g....

Awesome! It might be worth adding SCAD and MCP to the defaults because of their nice statistical properties (and it's hard to find these implemented penalties in python!)

Thanks for working on the bug in #1604! I suspect this bug is silently affecting people currently using torchmetrics. It looks like the current solution is a new argument `input_format`and...

Sorry I should have been more careful with my language! Let me rephrase “The auto option will always be buggy if the user inputs logits” to “The ‘auto’ option leaves...

Defaulting to `logits` might make the most sense. I suspect is that it's more common for users to default to outputting logits than outputting probabilities (e.g. https://pytorch.org/tutorials/beginner/blitz/cifar10_tutorial.html, https://github.com/huggingface/pytorch-image-models/blob/main/train.py, https://github.com/pytorch/examples/blob/main/imagenet/main.py) probably...

Looks like a similar thing happens in MultilabelAccuracy (https://torchmetrics.readthedocs.io/en/stable/classification/accuracy.html)

I installed version 1.2.dev0 from github get the same answers as you do above i.e. batch-gradient seems correct but there seems to be an issue with cdfast.