▎ tests/forecasting/bttf.py is the only forecasting test that makes no assertions — it computes an MSE and only logs it, so the test passes even if BTTF returns all-NaN output. Every sibling forecasting test asserts no-NaN output and uses a masked MSE.
▎
▎ BTTF is a Gibbs-sampling Bayesian model with no separate fit()/save step, so it can't get the full 5-method test suite the neural models have — but its single test should at least assert on its output like the others. I'd like to add a no-NaN assertion, pass the NaN mask to calc_mse, and assert the MSE is finite. Change is ready; PR against dev.
▎ tests/forecasting/bttf.py is the only forecasting test that makes no assertions — it computes an MSE and only logs it, so the test passes even if BTTF returns all-NaN output. Every sibling forecasting test asserts no-NaN output and uses a masked MSE.
▎
▎ BTTF is a Gibbs-sampling Bayesian model with no separate fit()/save step, so it can't get the full 5-method test suite the neural models have — but its single test should at least assert on its output like the others. I'd like to add a no-NaN assertion, pass the NaN mask to calc_mse, and assert the MSE is finite. Change is ready; PR against dev.