mindspore.ops.Adam

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

Updates gradients by the Adaptive Moment Estimation (Adam) algorithm.

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

For more details, please refer to mindspore.nn.Adam.

The updating formulas are as follows,

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

\(m\) 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 var, \(\epsilon\) 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:
  • var (Tensor) - Weights to be updated. The shape is \((N, *)\) where \(*\) means, any number of additional dimensions. The data type can be float16 or float32.

  • m (Tensor) - The 1st moment vector in the updating formula, the shape and data type value should be the same as var.

  • v (Tensor) - the 2nd moment vector in the updating formula, the shape and data type value should be the same as var. Mean square gradients with the same type as var.

  • beta1_power (float) - \(beta_1^t(\beta_1^{t})\) in the updating formula, the data type value should be the same as var.

  • beta2_power (float) - \(beta_2^t(\beta_2^{t})\) in the updating formula, the data type value should be the same as var.

  • lr (float) - \(l\) in the updating formula. The paper suggested value is \(10^{-8}\), the data type value should be the same as var.

  • beta1 (float) - The exponential decay rate for the 1st moment estimations, the data type value should be the same as var. The paper suggested value is \(0.9\)

  • beta2 (float) - The exponential decay rate for the 2nd moment estimations, the data type value should be the same as var. The paper suggested value is \(0.999\)

  • epsilon (float) - Term added to the denominator to improve numerical stability.

  • gradient (Tensor) - Gradient, has the same shape and data type as var.

Outputs:

Tuple of 3 Tensor, the updated parameters.

  • var (Tensor) - The same shape and data type as Inputs var.

  • m (Tensor) - The same shape and data type as Inputs m.

  • v (Tensor) - The same shape and data type as Inputs v.

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

  • TypeError – If var, m or v is not a Tensor.

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

Supported Platforms:

Ascend GPU CPU

Examples

>>> class Net(nn.Cell):
...     def __init__(self):
...         super(Net, self).__init__()
...         self.apply_adam = ops.Adam()
...         self.var = Parameter(Tensor(np.ones([2, 2]).astype(np.float32)), name="var")
...         self.m = Parameter(Tensor(np.ones([2, 2]).astype(np.float32)), name="m")
...         self.v = Parameter(Tensor(np.ones([2, 2]).astype(np.float32)), name="v")
...     def construct(self, beta1_power, beta2_power, lr, beta1, beta2, epsilon, grad):
...         out = self.apply_adam(self.var, self.m, self.v, beta1_power, beta2_power, lr, beta1, beta2,
...                               epsilon, grad)
...         return out
...
>>> net = Net()
>>> gradient = Tensor(np.ones([2, 2]).astype(np.float32))
>>> output = net(0.9, 0.999, 0.001, 0.9, 0.999, 1e-8, gradient)
>>> print(net.var.asnumpy())
[[0.9996838 0.9996838]
 [0.9996838 0.9996838]]