dagma.linear.DagmaLinear._adam_update¶

dagma.linear.DagmaLinear._adam_update(grad: numpy.ndarray, iter: int, beta_1: float, beta_2: float) → numpy.ndarray¶

Performs one update of Adam.

Parameters:¶

grad : np.ndarray¶

Current gradient of the objective.

iter : int¶

Current iteration number.

beta_1 : float¶

Adam hyperparameter.

beta_2 : float¶

Adam hyperparameter.

Returns:¶

Updates the gradient by the Adam method.

Return type:¶

np.ndarray