def test_polynomial_decay():
lr1 = dr.polynomial_decay_lr(learning_rate, end_learning_rate, total_step,
step_per_epoch, decay_epoch, power)
assert len(lr1) == total_step
lr2 = dr.polynomial_decay_lr(learning_rate, end_learning_rate, total_step,
step_per_epoch, decay_epoch, power, True)
assert len(lr2) == total_step
def test_learning_rate_value(self):
lr = -1.0
with pytest.raises(ValueError):
dr.exponential_decay_lr(lr, decay_rate, total_step, step_per_epoch, decay_epoch)
with pytest.raises(ValueError):
dr.polynomial_decay_lr(lr, end_learning_rate, total_step, step_per_epoch, decay_epoch, power)
def test_total_step1(self):
total_step1 = 2.0
with pytest.raises(ValueError):
dr.exponential_decay_lr(learning_rate, decay_rate, total_step1, step_per_epoch, decay_epoch)
with pytest.raises(ValueError):
dr.cosine_decay_lr(min_lr, max_lr, total_step1, step_per_epoch, decay_epoch)
with pytest.raises(ValueError):
dr.polynomial_decay_lr(learning_rate, end_learning_rate, total_step1, step_per_epoch, decay_epoch, power)
def test_decay_epoch1(self):
decay_epoch1 = 'm'
with pytest.raises(TypeError):
dr.exponential_decay_lr(learning_rate, decay_rate, total_step,
step_per_epoch, decay_epoch1)
with pytest.raises(TypeError):
dr.cosine_decay_lr(min_lr, max_lr, total_step, step_per_epoch,
decay_epoch1)
with pytest.raises(TypeError):
dr.polynomial_decay_lr(learning_rate, end_learning_rate,
total_step, step_per_epoch, decay_epoch1,
power)
def test_adam_group1():
""" test_adam_group_lr_and_weight_decay """
inputs = Tensor(np.ones([1, 64]).astype(np.float32))
label = Tensor(np.zeros([1, 10]).astype(np.float32))
net = Net()
net.set_train()
loss = nn.SoftmaxCrossEntropyWithLogits()
net_with_loss = WithLossCell(net, loss)
all_params = net.trainable_params()
poly_decay_lr = polynomial_decay_lr(0.01,
0.0001,
total_step=10,
step_per_epoch=1,
decay_epoch=3,
power=1.0)
group_params = [{
'params': [all_params[0]],
'lr': poly_decay_lr,
'weight_decay': 0.9
}, {
'params': [all_params[1]]
}]
optimizer = nn.Adam(group_params, learning_rate=0.1)
train_network = TrainOneStepCell(net_with_loss, optimizer)
_executor.compile(train_network, inputs, label)
def test_end_learning_rate_type(self):
lr = True
with pytest.raises(TypeError):
dr.polynomial_decay_lr(learning_rate, lr, total_step,
step_per_epoch, decay_epoch, power)
def test_update_decay_epoch(self):
update_decay_epoch = 1
with pytest.raises(TypeError):
dr.polynomial_decay_lr(learning_rate, end_learning_rate,
total_step, step_per_epoch, decay_epoch,
power, update_decay_epoch)
def test_power(self):
power1 = True
with pytest.raises(TypeError):
dr.polynomial_decay_lr(learning_rate, end_learning_rate,
total_step, step_per_epoch, decay_epoch,
power1)