提交 b6fa79b3 authored 作者: Frederic Bastien's avatar Frederic Bastien

lower the batch size of test_lenet_108 and test_lenet256 as the test was too long.

上级 c750c40d
......@@ -214,9 +214,8 @@ def test_conv_nnet2():
print rval_cpu[0], rval_gpu[0],rval_cpu[0]-rval_gpu[0]
assert numpy.allclose(rval_cpu, rval_gpu,rtol=1e-4,atol=1e-4)
def run_conv_nnet2_classif(shared_fn, isize, ksize):
def run_conv_nnet2_classif(shared_fn, isize, ksize, n_batch=60):
n_batch = 60
shape_img = (n_batch, 1, isize, isize)
n_kern = 20 # 6 were used in LeNet5
......@@ -286,7 +285,9 @@ def test_lenet_32(): #CIFAR10 / Shapeset
run_test_conv_nnet2_classif(23485, 32, 5)
def test_lenet_108(): # NORB
run_test_conv_nnet2_classif(23485, 108, 7)
#nbatch=10 as otherwise too slow for test
run_test_conv_nnet2_classif(23485, 108, 7, 10)
def test_lenet_256(): # ImageNet
run_test_conv_nnet2_classif(23485, 256, 9)
#nbatch=10 as otherwise too slow for test
run_test_conv_nnet2_classif(23485, 256, 9, 10)
Markdown 格式
0%
您添加了 0 到此讨论。请谨慎行事。
请先完成此评论的编辑!
注册 或者 后发表评论