This resembles to the babl issue we are having. Is it possible that this is hardware related?

Pierre Neidhardt <mail@ambrevar.xyz> ezt írta (időpont: 2020. febr. 11., Ke 16:19):
Tests fail with:

--8<---------------cut here---------------start------------->8---
=================================== FAILURES ===================================
__________________________________ test_selu ___________________________________

    def test_selu():
        x = K.placeholder(ndim=2)
        f = K.function([x], [activations.selu(x)])
        alpha = 1.6732632423543772848170429916717
        scale = 1.0507009873554804934193349852946

        positive_values = get_standard_values()
        result = f([positive_values])[0]
        assert_allclose(result, positive_values * scale, rtol=1e-05)

        negative_values = np.array([[-1, -2]], dtype=K.floatx())

        result = f([negative_values])[0]
        true_result = (np.exp(negative_values) - 1) * scale * alpha

>       assert_allclose(result, true_result)
E       AssertionError:
E       Not equal to tolerance rtol=1e-07, atol=0
E
E       Mismatch: 50%
E       Max absolute difference: 1.1920929e-07
E       Max relative difference: 1.0726715e-07
E        x: array([[-1.111331, -1.520167]], dtype=float32)
E        y: array([[-1.111331, -1.520167]], dtype=float32)

tests/keras/activations_test.py:226: AssertionError
--8<---------------cut here---------------end--------------->8---

See attached log.

--
Pierre Neidhardt
https://ambrevar.xyz/