at line 249 scale_cost is given to prepare and it just ignores it. normally it should've given in compute_batch_distances as the solver argument. This is bad since when I removed that line the tests still passed and when I gave the argument to the solver args it failed. This means gt was also computed without scale_cost but it was assumed to be.
Another bug found in https://github.com/theislab/moscot/pull/696/
https://github.com/theislab/moscot/blob/20c891a26bff4f7e11d72308ebb1f90b2610cb08/tests/problems/time/test_mixins.py#L238-L264
at line 249
scale_cost
is given to prepare and it just ignores it. normally it should've given incompute_batch_distances
as the solver argument. This is bad since when I removed that line the tests still passed and when I gave the argument to the solver args it failed. This means gt was also computed withoutscale_cost
but it was assumed to be.Fixed in this commit https://github.com/theislab/moscot/pull/696/commits/e1718c94a373219c48336a834a99868694ba4ede