Debug test reproducibility on MKL

Author: cwindolf

tests/test_drift_util.py

@@ -15,7 +15,7 @@ def test_shifted_waveforms():
     assert np.array_equal(np.unique(reg_geom[:, 1]), [1, 2, 3, 4, 5, 6, 7])
 
     # fixed check
-    waveforms = np.arange(15).reshape(5, 3)[:, None, :]
+    waveforms = np.arange(15).reshape(5, 3)[:, None, :].astype(np.float32)
     w = drift_util.get_waveforms_on_static_channels(waveforms, geom=geom)
     assert np.array_equal(w, waveforms)
     w = drift_util.get_waveforms_on_static_channels(

tests/test_matching.py

@@ -92,8 +92,8 @@ def test_tiny(tmp_path):
     assert np.isclose(
         torch.square(res["residual"]).mean(),
         0.0,
-        atol=1e-4,
     )
+    print(f"A {torch.square(res['conv']).mean()=}")
     assert np.isclose(
         torch.square(res["conv"]).mean(),
         0.0,
@@ -125,8 +125,8 @@ def test_tiny(tmp_path):
     assert np.isclose(
         torch.square(res["residual"]).mean(),
         0.0,
-        atol=1e-4,
     )
+    print(f"B {torch.square(res['conv']).mean()=}")
     assert np.isclose(
         torch.square(res["conv"]).mean(),
         0.0,
@@ -265,14 +265,13 @@ def test_tiny_up(tmp_path, up_factor=8):
     print(f'{np.c_[res["times_samples"], res["labels"], res["upsampling_indices"]]=}')
     print(f"{np.c_[times, labels, upsampling_indices]=}")
     print(f'{torch.square(res["residual"]).mean()=}')
-    print(f'{torch.square(res["conv"]).mean()=}')
+    print(f"C {torch.square(res['conv']).mean()=}")
     assert res["n_spikes"] == len(times)
     assert np.array_equal(res["times_samples"], times)
     assert np.array_equal(res["labels"], labels)
     assert np.isclose(
         torch.square(res["residual"]).mean(),
         0.0,
-        atol=1e-4,
     )
     assert np.isclose(
         torch.square(res["conv"]).mean(),
@@ -413,12 +412,12 @@ def static_tester(tmp_path, up_factor=1):
     assert np.isclose(
         torch.square(res["residual"]).mean(),
         0.0,
-        atol=1e-4,
     )
+    print(f"D {torch.square(res['conv']).mean()=}")
     assert np.isclose(
         torch.square(res["conv"]).mean(),
         0.0,
-        atol=1e-4,
+        atol=1e-3,
     )