def test_networkpass_multiple_ops(ops):
def prepare_input(v):
rng = np.random.RandomState(389)
v.d = rng.randn(*v.d.shape)
x = nn.Variable((4, 3))
prepare_input(x)
for n, ref_n in zip(nnp_get_topo(prepare_model(ops.s_module), 'left', ops.modifier(x)),
nnp_get_topo(prepare_model(ops.d_module), 'right')):
ref_inputs = list(ref_n.inputs.values())
ref_outputs = list(ref_n.outputs.values())
inputs = list(n.inputs.values())
outputs = list(n.outputs.values())
prepare_input(ref_inputs[0])
for ref_d, d in zip(forward_variable(ref_inputs, ref_outputs, 'left', True),
forward_variable(inputs, outputs, 'right', True)):
assert_allclose(d, ref_d)
def verify_equivalence(ref_v, v):
if isinstance(ref_v, legacy_nnp_graph.NnpNetwork):
ref_inputs = list(ref_v.inputs.values())
inputs = list(v.inputs.values())
ref_outputs = list(ref_v.outputs.values())
outputs = list(v.outputs.values())
for ref_d, d in zip(forward_variable(ref_inputs, ref_outputs, 'left'),
forward_variable(inputs, outputs, 'right')):
assert_allclose(ref_d, d)
assert_topology(ref_outputs, outputs)
elif isinstance(ref_v, nn.Variable):
compare_nn_variable_metadata(ref_v, v)
elif isinstance(ref_v, tuple):
compare_nn_variable_with_name(ref_v, v)
else:
print("compare {} <==> {}".format(ref_v, v))
assert ref_v == v
def load_model_from_nnp_graph_and_forward(nnp_file, batch_size):
nnp = nnp_graph.NnpLoader(nnp_file)
network = nnp.get_network(nnp.get_network_names()[0],
batch_size=batch_size)
inputs = list(network.inputs.values())
outputs = list(network.outputs.values())
out = []
for d in forward_variable(inputs, outputs, 'left'):
out.append(d)
return d