Éric Dupuis

As a temporary solution for git conflicts, maybe `nbdev` is worth exploring : https://nbdev.fast.ai/tutorials/git_friendly_jupyter.html I think it's nice that when writing a notebook, we don't have to develop tests, they're...

Oh, I thought that the content of notebooks was tested too. Thinking of it, it would be a mess if the output results are not hard-coded somewhere like we do...

So do we want to apply this approach to notebooks too? I would say that current notebooks are currently much more oriented towards "practical steps". But perhaps this comes quite...

closing, this has been dealt with in #1720

So I hijacked the PR a bit to test general benchmark tools. I needed to install ``pytest-benchmark``and ``pytest-benchmark[histogram]``. The idea would be to have a benchmarking suite, similar to the...

I did a test with EmpiricalQuantileMapping, 30 years, 3 locations, dayofyear-31. Sorting is still advantageous, but the effect is marginal. There is maybe an overhead which is just much greater...

It seems that removing `stack` on time dimensions and use `flatten/ravel` reduces computing time by about 25-30% when computing EmpiricalQuantileMapping! EDIT: Speed up is not clear, I was now trying...

See #1743 for additional tests. `_sortquantile` has performed very well, I think it's worth setting `allow_sortquantile=True` by default, which begs the question, do we need this switch now haha?

Concerning `vecquantiles` we have the same problem that we had for the old implementation of `quantiles` : if `time` has object values, it is very slow. However, in other cases,...

In the last implementation above, since I had no `njit` involved, I figured out it would be very similar to using `vectorize=True` in a `xr.apply_ufunc` ```python def vecquantiles(da, rnk, dim):...