def test_parameter_update_int32_and_tensor():
""" test_parameter_update """
net = Net()
loss = nn.SoftmaxCrossEntropyWithLogits()
optimizer = Momentum(net.get_parameters(), Tensor(np.array([0.1, 0.01, 0.001]), mstype.float32), 0.001)
net_with_loss = WithLossCell(net, loss)
train_network = TrainOneStepCell(net_with_loss, optimizer)
# compile train graph
train_network.set_train()
inputs = Tensor(np.ones([1, 64]).astype(np.float32))
label = Tensor(np.zeros([1, 10]).astype(np.float32))
_executor.compile(train_network, inputs, label)
# test tensor
param_lr = train_network.parameters_dict()['learning_rate']
update_network = ParameterUpdate(param_lr)
update_network.phase = 'update_param'
input_lr = Tensor(np.array([0.2, 0.02, 0.002]), mstype.float32)
_executor.compile(update_network, input_lr)
# test int32
param_step = train_network.parameters_dict()['global_step']
update_global_step = ParameterUpdate(param_step)
input_step = Tensor(np.array([1000]), mstype.int32)
_executor.compile(update_global_step, input_step)
def test_parameter_update_float32():
""" test_parameter_update """
net = Net()
loss = nn.SoftmaxCrossEntropyWithLogits()
optimizer = Momentum(net.get_parameters(), 0.01, 0.001)
net_with_loss = WithLossCell(net, loss)
train_network = TrainOneStepCell(net_with_loss, optimizer)
# compile train graph
train_network.set_train()
inputs = Tensor(np.ones([1, 64]).astype(np.float32))
label = Tensor(np.zeros([1, 10]).astype(np.float32))
_executor.compile(train_network, inputs, label)
# construct and compile update graph
param_lr = train_network.parameters_dict()['learning_rate']
update_network = ParameterUpdate(param_lr)
update_network.phase = 'update_param'
input_lr = Tensor(0.0001, mstype.float32)
_executor.compile(update_network, input_lr)