Closed szhan closed 1 month ago
I am still getting a few assertion fails for the test_ts_simple_n8
tests. They are caused by high relative differences but tiny absolute differences between log-likelihood values and FB product matrix values.
For example:
E AssertionError:
E Not equal to tolerance rtol=1e-09, atol=0
E
E Mismatched elements: 1 / 1 (100%)
E Max absolute difference: 8.19678347e-16
E Max relative difference: 1.88888889
E x: array(-1.253626e-15)
E y: array(-4.339474e-16)
And
E AssertionError:
E Not equal to tolerance rtol=1e-09, atol=0
E
E Mismatched elements: 25 / 27 (92.6%)
E Max absolute difference: 9.64327467e-17
E Max relative difference: 1.
E x: array([9.643275e-17, 1.928655e-16, 1.928655e-16, 1.928655e-16,
E 1.928655e-16, 1.928655e-16, 1.928655e-16, 1.928655e-16,
E 1.928655e-16, 1.928655e-16, 1.928655e-16, 1.928655e-16,...
E y: array([9.643275e-17, 9.643275e-17, 9.643275e-17, 9.643275e-17,
E 9.643275e-17, 9.643275e-17, 9.643275e-17, 9.643275e-17,
E 9.643275e-17, 9.643275e-17, 9.643275e-17, 9.643275e-17,...
Assert allclose at rtol=1e-09
and atol=0
may be too strict.
Perhaps we could use numpy.testing.assert_almost_equal
instead. Doc here.
"The test verifies that the elements of actual and desired satisfy. abs(desired-actual) < float64(1.5 * 10**(-decimal))"
It's strange that four tests failed on my machine but passed during CI.
Not sure if this is due to architecture. I'm using Apple Silicon. I guess the workflows here use an Intel chip?
Fix #39