mindspore.ops.AdamNoUpdateParam

class mindspore.ops.AdamNoUpdateParam(use_locking=False, use_nesterov=False)[source]

Updates gradients by the Adaptive Moment Estimation (Adam) algorithm. This operator do not update the parameter, but calculate the value that should be added to the parameter instead.

The Adam algorithm is proposed in Adam: A Method for Stochastic Optimization.

The updating formulas are as follows,

\[\begin{split}\begin{array}{ll} \\ m = \beta_1 * m + (1 - \beta_1) * g \\ v = \beta_2 * v + (1 - \beta_2) * g * g \\ l = \alpha * \frac{\sqrt{1-\beta_2^t}}{1-\beta_1^t} \\ \Delta{w} = - l * \frac{m}{\sqrt{v} + \epsilon} \end{array}\end{split}\]

\(m\) represents the 1st moment vector, \(v\) represents the 2nd moment vector, \(g\) represents gradient, \(l\) represents scaling factor lr, \(\beta_1, \beta_2\) represent beta1 and beta2, \(t\) represents updating step while \(beta_1^t(\beta_1^{t})\) and \(beta_2^t(\beta_2^{t})\) represent beta1_power and beta2_power, \(\alpha\) represents learning_rate, \(w\) represents the parameter to be updated, \(\epsilon\) represents epsilon.

Parameters

  • use_locking (bool) – Whether to enable a lock to protect variable tensors from being updated. If true, updates of the var, m, and v tensors will be protected by a lock. If false, the result is unpredictable. Default: False.

  • use_nesterov (bool) – Whether to use Nesterov Accelerated Gradient (NAG) algorithm to update the gradients. If true, update the gradients using NAG. If false, update the gradients without using NAG. Default: False.

Inputs:

  • m (Tensor) - The 1st moment vector in the updating formula. The shape is \((N, *)\) where \(*\) means, any number of additional dimensions. The data type must be float32.

  • v (Tensor) - the 2nd moment vector in the updating formula. The shape must be the same as m. The data type must be float32.

  • beta1_power (Tensor) - \(beta_1^t(\beta_1^{t})\) in the updating formula. The shape is \((1, )\) and the data type must be float32.

  • beta2_power (Tensor) - \(beta_2^t(\beta_2^{t})\) in the updating formula. The shape is \((1, )\) and the data type must be float32.

  • lr (Tensor) - \(l\) in the updating formula. The shape is \((1, )\) and the data type must be float32.

  • beta1 (Tensor) - The exponential decay rate for the 1st moment estimations. The shape is \((1, )\) and the data type must be float32.

  • beta2 (Tensor) - The exponential decay rate for the 2nd moment estimations. The shape is \((1, )\) and the data type must be float32.

  • epsilon (Tensor) - Term added to the denominator to improve numerical stability. The shape is \((1, )\) and the data type must be float32.

  • gradient (Tensor) - Gradient, the shape must be the same as m, the data type must be float32.

Outputs:

Tensor, whose shape and data type are the same with Inputs gradient, is a value that should be added to the parameter to be updated.

Raises

  • TypeError – If neither use_locking nor use_nesterov is a bool.

  • TypeError – If m, v, beta1_power, beta2_power1, lr, beta1, beta2, epsilon or gradient is not a Tensor.

Supported Platforms:

CPU

Examples

>>> class Net(nn.Cell):
...     def __init__(self):
...         super(Net, self).__init__()
...         self.adam = ops.AdamNoUpdateParam()
...         self.m = Parameter(Tensor(np.array([[0.1, 0.1, 0.1], [0.2, 0.2, 0.2]]).astype(np.float32)),
...                            name="m")
...         self.v = Parameter(Tensor(np.array([[0.1, 0.1, 0.1], [0.2, 0.2, 0.2]]).astype(np.float32)),
...                            name="v")
...     def construct(self, beta1_power, beta2_power, lr, beta1, beta2, epsilon, grad):
...         out = self.adam(self.m, self.v, beta1_power, beta2_power, lr, beta1, beta2, epsilon, grad)
...         return out
>>> net = Net()
>>> beta1_power = Tensor(0.9, ms.float32)
>>> beta2_power = Tensor(0.999, ms.float32)
>>> lr = Tensor(0.001, ms.float32)
>>> beta1 = Tensor(0.9, ms.float32)
>>> beta2 = Tensor(0.999, ms.float32)
>>> epsilon = Tensor(1e-8, ms.float32)
>>> gradient = Tensor(np.array([[0.1, 0.1, 0.1], [0.1, 0.1, 0.1]]).astype(np.float32))
>>> result = net(beta1_power, beta2_power, lr, beta1, beta2, epsilon, gradient)
>>> print(result)
[[-0.00010004 -0.00010004 -0.00010004]
[-0.00013441 -0.00013441 -0.00013441]]