[GitHub] lebeg commented on a change in pull request #12427: Improved test for BilinearSampler

[GitBox <gi...@apache.org>]

lebeg commented on a change in pull request #12427: Improved test for BilinearSampler
URL: https://github.com/apache/incubator-mxnet/pull/12427#discussion_r214615370
 
 

 ##########
 File path: tests/python/gpu/test_operator_gpu.py
 ##########
 @@ -1918,6 +1918,65 @@ def test_softmax_activation():
         assert_almost_equal(cpu_a.grad.asnumpy(), gpu_a.grad.asnumpy(),
                 atol = 1e-3, rtol = 1e-3)
 
+
+@with_seed()
+def test_bilinear_sampler_versions():
+    data = mx.sym.Variable('data')
+    grid = mx.sym.Variable('grid')
+    sym1 = mx.sym.BilinearSampler(data=data, grid=grid)
+    sym2 = mx.sym.BilinearSampler(data=data, grid=grid, cudnn_off=True)
+    sym3 = mx.sym.BilinearSampler(data=data, grid=grid)
+
+    test_cases = [[(1,3,15,16),(1,2,10,10)],
+                 [(1,6,7,16),(1,2,10,4)],
+                 [(1,7,3,16),(1,2,8,11)],
+                 [(1,9,50,50),(1,2,50,50)]]
+
+    for item in test_cases:
+        data_shape, grid_shape = item
+        # kWriteTo
+        exe_cpu = sym1.simple_bind(data=data_shape, grid=grid_shape, ctx=mx.cpu(), grad_req='write')
+        exe_gpu = sym2.simple_bind(data=data_shape, grid=grid_shape, ctx=default_context(), grad_req='write')
+        exe_cudnn = sym3.simple_bind(data=data_shape, grid=grid_shape, ctx=default_context(), grad_req='write')
+        exe_list = [exe_cpu, exe_gpu, exe_cudnn]
+        ref_idx = 0
+        test_data = np.random.uniform(low=-0.1, high=0.1,size=data_shape).astype(np.float32)
+        test_grid = np.random.uniform(low=-2, high=2, size=grid_shape).astype(np.float32)
+        for exe in exe_list:
+            exe.arg_dict['data'][:] = test_data
+            exe.arg_dict['grid'][:] = test_grid
+            exe.forward(is_train=True)
+            assert_almost_equal(exe_list[0].outputs[0].asnumpy(), exe.outputs[0].asnumpy(), rtol=1e-3, atol=1e-5)
 
 Review comment:
   The NumPy documentation suggests [assert_allclose](https://docs.scipy.org/doc/numpy/reference/generated/numpy.testing.assert_allclose.html#numpy.testing.assert_allclose) as better alternative to [assert_almost_equal](https://docs.scipy.org/doc/numpy/reference/generated/numpy.testing.assert_almost_equal.html).

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services