testing different strides and pooling modes

theano/sandbox/cuda/tests/test_dnn.py

@@ -6,6 +6,8 @@ import theano
 from theano.gof.python25 import any
 import theano.tensor as T
 import theano.tests.unittest_tools as utt
+from theano.sandbox.neighbours import images2neibs, neibs2images
+
 
 # Skip test if cuda_ndarray is not available.
 import theano.sandbox.cuda as cuda
@@ -21,30 +23,46 @@ else:
     mode_without_gpu = theano.compile.mode.get_default_mode().excluding('gpu')
 
 
+
+def pool_2d_i2n(input, ds=(2, 2), strides=None, pool_function=T.max, mode='ignore_borders'):
+    if strides is None:
+        strides = ds
+
+    if strides[0] > ds[0] or strides[1] > ds[1]:
+        raise RuntimeError("strides should be smaller than or equal to ds, strides=(%d, %d) and ds=(%d, %d)" %
+            (strides + ds))
+
+    shape = input.shape
+    neibs = images2neibs(input, ds, strides, mode=mode)
+    pooled_neibs = pool_function(neibs, axis=1)
+
+    output_width = (shape[2] - ds[0]) // strides[0] + 1
+    output_height = (shape[3] - ds[1]) // strides[1] + 1
+
+    pooled_output = pooled_neibs.reshape((shape[0], shape[1], output_width, output_height))
+    return pooled_output
+
+
 def test_pooling():
     if not cuda.dnn.dnn_available():
         raise SkipTest(cuda.dnn.dnn_available.msg)
     
     x = T.tensor4()
 
-    out1 = cuda.dnn.dnn_pool(x, ws=(5,5), stride=(5, 5))
-    out2 = T.signal.downsample.max_pool_2d(x, (5, 5))
-
-    f1 = theano.function([x], out1)
-    f2 = theano.function([x], out2)
-
-    data = numpy.random.normal(0, 1, (1, 10, 100, 100)).astype("float32")
-    a = f1(data).__array__()
-
-    b = f2(data).__array__()
-
-    assert numpy.allclose(a, b)
+    for func in (T.max, T.mean):
+        for ws in (4, 5):
+            for stride in (2, 3):
+                out1 = cuda.dnn.dnn_pool(x, ws=(ws, ws), stride=(stride, stride),
+                    mode='max' if func is T.max else "average")
+                out2 = pool_2d_i2n(x, ds=(ws, ws), strides=(stride, stride),
+                    pool_function=func)
 
-    gf1 = theano.function([x], theano.grad(out1.sum(), x))
-    gf2 = theano.function([x], theano.grad(out2.sum(), x))
+                f1 = theano.function([x], out1)
+                f2 = theano.function([x], out2)
 
-    ga = gf1(data).__array__()
+                data = numpy.random.normal(0, 1, (1, 10, 100, 100)).astype("float32")
+                a = f1(data).__array__()
 
-    gb = gf2(data).__array__()
+                b = f2(data).__array__()
 
-    assert numpy.allclose(ga, gb)
+                assert numpy.allclose(a, b, atol=numpy.finfo(numpy.float32).eps)