def build_test_sequential_graph(self, backend):
"""Create a sequential model."""
np_dtype = test_backend_dtypes[backend]
self.expected_dtype = datatypes.np_to_smaug_type[np_dtype]
with Graph(name="test_sequential_graph", backend=backend) as graph:
input_tensor = Tensor(
data_layout=types_pb2.NCHW,
tensor_data=np.random.rand(1, 3, 28, 28).astype(np_dtype))
filter_tensor0 = Tensor(
data_layout=types_pb2.NCHW,
tensor_data=np.random.rand(64, 3, 3, 3).astype(np_dtype))
filter_tensor1 = Tensor(
data_layout=types_pb2.NCHW,
tensor_data=np.random.rand(64, 64, 3, 3).astype(np_dtype))
weight_tensor0 = Tensor(
data_layout=types_pb2.NC,
tensor_data=np.random.rand(254, 12544).astype(np_dtype))
weight_tensor1 = Tensor(
data_layout=types_pb2.NC,
tensor_data=np.random.rand(10, 254).astype(np_dtype))
bn_mean_tensor = Tensor(
data_layout=types_pb2.NC,
tensor_data=np.random.rand(1, 64).astype(np_dtype))
bn_var_tensor = Tensor(
data_layout=types_pb2.NC,
tensor_data=np.random.rand(1, 64).astype(np_dtype))
bn_gamma_tensor = Tensor(
data_layout=types_pb2.NC,
tensor_data=np.random.rand(1, 64).astype(np_dtype))
bn_beta_tensor = Tensor(
data_layout=types_pb2.NC,
tensor_data=np.random.rand(1, 64).astype(np_dtype))
out = data_op.input_data(input_tensor, "input")
out = nn_ops.convolution(
out, filter_tensor0, stride=[1, 1], padding="same", name="conv0")
out = activation_ops.relu(out, "conv0_relu")
out = nn_ops.batch_norm(
out, bn_mean_tensor, bn_var_tensor, bn_gamma_tensor, bn_beta_tensor,
name="bn")
out = nn_ops.convolution(
out, filter_tensor1, stride=[1, 1], padding="same", name="conv1")
out = activation_ops.relu(out, "conv1_relu")
out = nn_ops.max_pool(out, pool_size=[2, 2], stride=[2, 2], name="pool")
out = array_ops.flatten(out, "flatten")
out = nn_ops.mat_mul(out, weight_tensor0, name="fc0")
out = activation_ops.relu(out, "fc0_relu")
out = nn_ops.mat_mul(out, weight_tensor1, name="fc1")
out = array_ops.expand_dims(out, 1, "expand_dims")
out = array_ops.squeeze(out, 1, "squeeze")
out = array_ops.reshape(out, [2, 5], types_pb2.NC, "reshape")
out = array_ops.repeat(out, [4, 2], "repeat")
out = array_ops.stack(out, 4, 1, "stack")
out0, out1, out2, out3 = array_ops.unstack(out, 1, "unstack")
out0 = array_ops.reshape(out0, [1, 1, 8, 10], types_pb2.NCHW, "reshape")
out0 = array_ops.padding(out0, [0, 0, 0, 0, 1, 1, 1, 1], "padding")
self.test_graph, _ = graph.to_proto()
self.backend = backend
self.alignment = global_vars.backend_alignment[backend]
def build_test_residual_graph(self, backend):
"""Create a residual model.
The graph contains a residual connection, where the output of conv0 and
conv2 is added at the end."""
np_dtype = test_backend_dtypes[backend]
self.expected_dtype = datatypes.np_to_smaug_type[np_dtype]
with Graph(name="test_residual_graph", backend=backend) as graph:
input_tensor = Tensor(
data_layout=types_pb2.NCHW,
tensor_data=np.random.rand(1, 1, 28, 28).astype(np_dtype))
filter_tensor0 = Tensor(
data_layout=types_pb2.NCHW,
tensor_data=np.random.rand(64, 1, 3, 3).astype(np_dtype))
filter_tensor1 = Tensor(
data_layout=types_pb2.NCHW,
tensor_data=np.random.rand(64, 1, 3, 3).astype(np_dtype))
filter_tensor2 = Tensor(
data_layout=types_pb2.NCHW,
tensor_data=np.random.rand(64, 64, 3, 3).astype(np_dtype))
bn_mean_tensor = Tensor(
data_layout=types_pb2.NC,
tensor_data=np.random.rand(1, 64).astype(np_dtype))
bn_var_tensor = Tensor(
data_layout=types_pb2.NC,
tensor_data=np.random.rand(1, 64).astype(np_dtype))
bn_gamma_tensor = Tensor(
data_layout=types_pb2.NC,
tensor_data=np.random.rand(1, 64).astype(np_dtype))
bn_beta_tensor = Tensor(
data_layout=types_pb2.NC,
tensor_data=np.random.rand(1, 64).astype(np_dtype))
act = data_op.input_data(input_tensor, "input")
x = nn_ops.convolution(
act, filter_tensor0, stride=[1, 1], padding="same", name="conv0")
out = nn_ops.convolution(
act, filter_tensor1, stride=[1, 1], padding="same", name="conv1")
out = nn_ops.batch_norm(
out, bn_mean_tensor, bn_var_tensor, bn_gamma_tensor, bn_beta_tensor,
name="bn")
out = activation_ops.relu(out, "relu")
out = nn_ops.convolution(
out, filter_tensor2, stride=[1, 1], padding="same", name="conv2")
out = math_ops.add(x, out, "add")
out = math_ops.mul(x, out, "mul")
# Concatenate the channel dimension of x and out.
axis = 1 if out.shape.layout == types_pb2.NCHW else 3
out = array_ops.concat([x, out], axis, "concat")
# Evenly split the tensor into 4 over the channel dimension.
out0, out1, out2, out3 = array_ops.split(out, 4, axis, "split")
out = math_ops.mul(
math_ops.add(out0, out1, "add1"), math_ops.add(out2, out3, "add2"),
"mul1")
self.test_graph, _ = graph.to_proto()
self.backend = backend
self.alignment = global_vars.backend_alignment[
self.test_graph.backend]
def test_activation_functions(self):
"""Test activation function attributes."""
with Graph("test_graph", "SMV") as test_graph:
tensor_data = np.random.rand(*self.tensor_shape).astype(np.float16)
input_tensor = Tensor(data_layout=types_pb2.NHWC,
tensor_data=tensor_data)
act = data_op.input_data(input_tensor, "input")
act = activation_ops.relu(act, "relu")
act = activation_ops.lrelu(act, slope=0.5, name="lrelu")
act = activation_ops.elu(act, alpha=0.2, name="elu")
act = activation_ops.selu(act,
alpha=0.4,
lambda_param=0.8,
name="selu")
act = activation_ops.tanh(act, "tanh")
act = activation_ops.hard_tanh(act,
min=-1.5,
max=1.5,
name="hard_tanh")
act = activation_ops.sigmoid(act, "sigmoid")
# Softmax expects NC format, so reorder NHWC to NC.
act = array_ops.reorder(act,
target_layout=types_pb2.NC,
name="reorder")
act = activation_ops.softmax(act, "softmax")
# ReLU
self.do_basic_test(test_graph, "relu", types_pb2.ReLU)
# LReLU
node = self.do_basic_test(test_graph, "lrelu", types_pb2.LReLU)
self.assertAlmostEqual(node.params.act_params.lrelu_params.slope, 0.5)
# ELU
node = self.do_basic_test(test_graph, "elu", types_pb2.ELU)
self.assertAlmostEqual(node.params.act_params.elu_params.alpha, 0.2)
# SELU
node = self.do_basic_test(test_graph, "selu", types_pb2.SELU)
self.assertAlmostEqual(node.params.act_params.elu_params.alpha, 0.4)
self.assertAlmostEqual(node.params.act_params.elu_params.lambda_param,
0.8)
# Tanh
self.do_basic_test(test_graph, "tanh", types_pb2.Tanh)
# HardTanh
node = self.do_basic_test(test_graph, "hard_tanh", types_pb2.HardTanh)
self.assertAlmostEqual(node.params.act_params.hard_tanh_params.min,
-1.5)
self.assertAlmostEqual(node.params.act_params.hard_tanh_params.max,
1.5)
# Sigmoid
self.do_basic_test(test_graph, "sigmoid", types_pb2.Sigmoid)
# Softmax
self.do_basic_test(test_graph,
"softmax",
types_pb2.Softmax,
tensor_shape=[2, 32768])