def test_reduce_front_sum_empty_batch(self, num_reduce_dim, seed, gc, dc):
np.random.seed(seed)
X = np.random.rand(0, 4, 3, 5).astype(np.float32)
def ref_sum(X):
return [np.sum(X, axis=(tuple(range(num_reduce_dim))))]
self.reduce_op_test("ReduceFrontSum", ref_sum, [X], ["input"],
num_reduce_dim, gc)
self.grad_variant_input_test("ReduceFrontSumGradient", X, ref_sum,
num_reduce_dim)
# test the second iteration
not_empty_X = np.random.rand(2, 4, 3, 5).astype(np.float32)
net = core.Net('test')
with core.DeviceScope(gc):
net.ReduceFrontSum(['X'], ['output'],
num_reduce_dim=num_reduce_dim)
workspace.CreateNet(net)
workspace.FeedBlob('X', not_empty_X)
workspace.RunNet(workspace.GetNetName(net))
output = workspace.FetchBlob('output')
np.testing.assert_allclose(output,
ref_sum(not_empty_X)[0],
atol=1e-3)
workspace.FeedBlob('X', X)
workspace.RunNet(workspace.GetNetName(net))
output = workspace.FetchBlob('output')
np.testing.assert_allclose(output, ref_sum(X)[0], atol=1e-3)
net_def.ParseFromString(f.read())
net_def.device_option.CopyFrom(device_opts)
C.create_net(net_def.SerializeToString())
C.feed_blob('data', img, device_opts.SerializeToString())
### Debug code
"""
for b in workspace.Blobs():
x = workspace.FetchBlob(b)
if type(x) != str:
print(str(b) + ': ' + str(x.shape))
"""
### End of debug code
print('Running net ' + workspace.GetNetName(net_def) + '...')
C.run_net(workspace.GetNetName(net_def), 1, False)
# Turn it into something we can play with and examine which is in a multi-dimensional array
results = workspace.FetchBlob('prob')
#print("results shape: ", results.shape)
# Quick way to get the top-1 prediction result
# Squeeze out the unnecessary axis. This returns a 1-D array of length 1000
preds = np.squeeze(results)
# Get the prediction and the confidence by finding the maximum value and index of maximum value in preds array
curr_pred, curr_conf = max(enumerate(preds), key=operator.itemgetter(1))
print("Prediction: ", curr_pred)
print("Confidence: ", curr_conf)
# Create GPU device option
device_opts = caffe2_pb2.DeviceOption()
if use_gpu == 0:
device_opts.device_type = caffe2_pb2.CPU
print('Running on CPU')
else:
device_opts.device_type = caffe2_pb2.HIP
device_opts.hip_gpu_id = 0
print('Running on HIP')
if use_gpu == 2:
engine_list = ['MIOPEN', '']
C.set_global_engine_pref({caffe2_pb2.HIP : engine_list})
print('Using MIOPEN')
C.feed_blob('data', img, device_opts.SerializeToString())
init_def = caffe2_pb2.NetDef()
with open(INIT_NET, 'rb') as f:
init_def.ParseFromString(f.read())
init_def.device_option.CopyFrom(device_opts)
C.run_net_once(init_def.SerializeToString())
net_def = caffe2_pb2.NetDef()
with open(PREDICT_NET, 'rb') as f:
net_def.ParseFromString(f.read())
net_def.device_option.CopyFrom(device_opts)
C.create_net(net_def.SerializeToString())
C.feed_blob('data', img, device_opts.SerializeToString())
C.benchmark_net(workspace.GetNetName(net_def), 10, num_iter, per_layer_timing)