#70: pull in pressiolinalg pypi package

[cwschilly]

Fixes #70

Related to pressio-linalg PR 8

---

Changes

pressio-linalg is now a required dependency.

• If you pass an MPI communicator to any of the functions, they will run in parallel. Otherwise, the library defaults to running serial.

---

Usage

import pressiolinalg.linalg [as pla]

Note that there is no hyphen for the import.

---

Current API

pla.max(a, axis=None, comm=None)
pla.min(a, axis=None, comm=None)
pla.mean(a, axis=None, dtype=None, comm=None)
pla.std(a, axis=None, dtype=None, ddof=0, comm=None)
pla.product(flag1, flag2, alpha, A, B, beta, C, comm=None)
pla.thin_svd(snapshots, comm=None, method='auto')

These are all subject to change and will be documented in this issue.

[cwschilly]

@fnrizzi @eparish1 This PR is ready for review.

I added a unit test for the new package, but it will likey become unnecessary once we begin to incorporate the package into source code (since existing tests would then catch any problems).

pressio-linalg also has its own testing suite already.

That being said, should I keep the test in or drop it from the PR?

[fnrizzi]

That being said, should I keep the test in or drop it from the PR?

the linalg library should have its own testsuite. since that is the case, we should remove tests related to it from this repo.
The romtools repo should only use the linalg package, so just assuming it works.

[fnrizzi]

@cwschilly can you also please replace in romtools all functions to call the linalg library functions?

[eparish1]

@cwschilly @fnrizzi I agree about Francesco's comments on the tests. But this looks good. I'm fine to replace the calls to linalg in this PR or another one, whatever is easiest.

[cwschilly]

@fnrizzi @eparish1 I've added in pla where applicable--there's a call to np.max in the Scaler that I couldn't update yet since the pla.max only works on rank-1 arrays.

[fnrizzi]

@fnrizzi @eparish1 I've added in pla where applicable--there's a call to np.max in the Scaler that I couldn't update yet since the pla.max only works on rank-1 arrays.

please add this to linalg then

[cwschilly]

@fnrizzi @eparish1 All calls to np.max, np.min and SVD (method of snapshots) have been replaced

[eparish1]

@cwschilly This is wonderful, thanks for doing all of this. This looks great by me. @fnrizzi good by you?

[eparish1]

@cwschilly @fnrizzi you both are tracking where things are at here better than me. Could one of you do the merge when things are ready? Thanks!

[cwschilly]

@fnrizzi @eparish1 This PR should be ready to go now