Having had a look around some more, there are a few more instances of svd_interface() that require passing random_state

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 88.13%. Comparing base (983e497) to head (92adc7f).
⚠️ Report is 6 commits behind head on main.

@@            Coverage Diff             @@
##             main     #616      +/-   ##
==========================================
+ Coverage   88.07%   88.13%   +0.05%     
==========================================
  Files         132      132              
  Lines        7943     7941       -2     
==========================================
+ Hits         6996     6999       +3     
+ Misses        947      942       -5     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

Added all other instances where svd_interface() is called, including situations where a random_state needed to be specified. I added random_state to any enclosing function that also takes the svd argument and which the documentation claims can have any value present in tensorly.SVD_FUNS, which includes randomized_svd potentially requiring random_state.

As a result, the following enclosing functions were modified to take random_state with default value set to None:

• _compute_projections() in _parafac2.py
• tensor_ring() in _tr_svd.py
• tensor_train() in _tt.py
• svd_compress_tensor_slices() in preprocessing.py

Additionally, I modified the svd_interface() calls in test_backend.py as one of those tests explicitly checks for svd=="randomized_svd" without passing a random_state, but perhaps this was intentional?

Take a look at the test errors here. I'm happy to discuss them if it's not clear what broke.