def __init__(self, learning_rate, end_learning_rate, warmup_steps, decay_steps, power): super(BertLearningRate, self).__init__() self.warmup_lr = WarmUpLR(learning_rate, warmup_steps) self.decay_lr = PolynomialDecayLR(learning_rate, end_learning_rate, decay_steps, power) self.warmup_steps = Tensor(np.array([warmup_steps]).astype(np.float32)) self.greater = P.Greater() self.one = Tensor(np.array([1.0]).astype(np.float32)) self.cast = P.Cast()
def __init__(self, learning_rate, warmup_steps, multi_epochs, steps_per_epoch, factor=10): super(CenterNetMultiEpochsDecayLR, self).__init__() self.warmup_flag = False if warmup_steps > 0: self.warmup_flag = True self.warmup_lr = WarmUpLR(learning_rate, warmup_steps) self.decay_lr = MultiEpochsDecayLR(learning_rate, multi_epochs, steps_per_epoch, factor) self.warmup_steps = Tensor(np.array([warmup_steps]).astype(np.float32)) self.greater = ops.Greater() self.one = Tensor(np.array([1.0]).astype(np.float32)) self.cast = ops.Cast()
def __init__(self, learning_rate, end_learning_rate, warmup_steps, decay_steps, power): super(CenterNetPolynomialDecayLR, self).__init__() self.warmup_flag = False if warmup_steps > 0: self.warmup_flag = True self.warmup_lr = WarmUpLR(learning_rate, warmup_steps) self.decay_lr = PolynomialDecayLR(learning_rate, end_learning_rate, decay_steps, power) self.warmup_steps = Tensor(np.array([warmup_steps]).astype(np.float32)) self.greater = ops.Greater() self.one = Tensor(np.array([1.0]).astype(np.float32)) self.cast = ops.Cast()
def __init__(self, learning_rate, end_learning_rate, warmup_steps, decay_steps, power): super(BertLearningRate, self).__init__() self.warmup_flag = False if warmup_steps > 0: self.warmup_flag = True self.warmup_lr = WarmUpLR(learning_rate, warmup_steps) self.decay_lr = PolynomialDecayLR(learning_rate, end_learning_rate, decay_steps, power) self.warmup_steps = ts.array([warmup_steps], dtype=ts.float32) self.greater = P.Greater() self.one = ts.array([1.0], dtype=ts.float32) self.cast = P.Cast()
def __init__(self, learning_rate, end_learning_rate, warmup_steps, decay_steps, power=1.0, use_cosine=True): super(LearningRate, self).__init__() self.warmup_flag = False if warmup_steps > 0: self.warmup_flag = True self.warmup_lr = WarmUpLR(learning_rate, warmup_steps) self.decay_lr = PolynomialDecayLR(learning_rate, end_learning_rate, decay_steps, power) self.cosine_decay_lr = CosineDecayLR(end_learning_rate, learning_rate, decay_steps) self.warmup_steps = Tensor(np.array([warmup_steps]).astype(np.float32)) self.greater = P.Greater() self.one = Tensor(np.array([1.0]).astype(np.float32)) self.cast = P.Cast() self.use_cosine = use_cosine