search

pytorch / contrib

Implementations of ideas from recent papers
391 stars 42 forks source link

SWA test broken #38

Open Balandat opened 3 years ago

Balandat commented 3 years ago

test_swa fails with the following: 

test/test_swa.py::TestSWA::test_swa
  /Users/balandat/Code/contrib/torchcontrib/optim/swa.py:200: UserWarning: SWA wasn't applied to param tensor([[-1.3171e-01,  3.3898e-01,  3.9688e-01,  9.4115e-01, -1.0891e-01],
          [-6.1716e-01, -9.9478e-02, -3.5716e-01,  6.9384e-01, -2.1573e-01],
          [-9.2800e-01,  2.5287e-01, -9.5098e-01, -3.5055e-01,  6.5104e-01],
          [ 6.6169e-01,  7.8536e-01,  6.6425e-01, -1.6037e+00, -6.1215e-01],
          [ 5.0540e-01, -8.8359e-01,  9.3367e-01, -3.3240e-01,  3.8069e-01],
          [ 1.0775e-01,  3.0351e-01,  1.3170e+00, -5.3968e-01,  1.8907e-01],
          [-4.6874e-02,  9.1908e-01, -1.7464e+00, -7.5991e-01, -4.2205e-01],
          [ 1.0197e+00,  1.0610e-01,  8.9256e-01,  5.9761e-01, -1.9891e-03],
          [ 3.0417e+00, -8.8183e-01, -4.5786e-01, -6.0704e-01,  3.2223e-01],
          [-1.6168e+00,  9.2239e-01, -2.0307e-01, -1.2542e-01,  1.5070e+00]],
         requires_grad=True); skipping it
    warnings.warn(

test/test_swa.py::TestSWA::test_swa
  /Users/balandat/Code/contrib/torchcontrib/optim/swa.py:200: UserWarning: SWA wasn't applied to param tensor([ 0.5962, -0.0670,  0.4809, -1.0233, -1.6968,  0.0400,  0.7110, -0.4441,
          -0.3767,  0.0970], requires_grad=True); skipping it
    warnings.warn(

-- Docs: https://docs.pytest.org/en/stable/warnings.html
================================================================================================================== short test summary info ===================================================================================================================
FAILED test/test_swa.py::TestSWA::test_swa - AssertionError: tensor(0.1980, grad_fn=<MaxBackward1>) not less than or equal to 1e-05 :