mojolearn.RMSprop

class mojolearn.RMSprop(params, lr=0.01, alpha=0.99, eps=1e-08, weight_decay=0.0, momentum=0.0, centered=False, numeric_mode=None)

Bases: _SeqOptimizer

RMSprop optimizer, as torch.optim.RMSprop.

Keeps a running average of squared gradients, v = alpha * v + (1 - alpha) * g**2, and divides the gradient by sqrt(v) + eps. With centered=True the squared running mean of the gradient is subtracted from v; with momentum > 0 a momentum buffer accumulates the scaled gradient. Defaults are torch’s.

Parameters:

• params (ndarray or list of ndarray) – Parameters to update in place.

• lr (float, default 1e-2) – Learning rate.

• alpha (float, default 0.99) – Smoothing constant of the squared-gradient average.

• eps (float, default 1e-8) – Term added to the denominator.

• weight_decay (float, default 0.0) – L2 penalty added to the gradient.

• momentum (float, default 0.0) – Momentum factor.

• centered (bool, default False) – Normalize by the estimated gradient variance.

• numeric_mode ({‘identical’, ‘deterministic’, ‘fast’} or None, default None) – Numeric mode of the update. None uses the library default.

Variables:

• t (int) – Number of steps taken.

• square_avg (ndarray) – Running average of squared gradients, flat over all parameters.

Notes

params must be C-contiguous float32 NumPy arrays, updated in place; anything else raises TypeError. step(grads) takes one gradient per parameter, in the same order and shapes; float64 gradients raise TypeError. lr is read at every step, so you can change opt.lr between steps, or set opt.lr_schedule to a schedule (StepLR, ExponentialLR, OneCycleLR, or any object with a one-based lr_at(t)), which sets lr before each step. state_dict() and load_state_dict() save and restore the step count, learning rate and state buffers.

Not supported: parameter groups, maximize, foreach/fused/capturable/ differentiable, sparse gradients and float64 parameters.

Examples

>>> import numpy as np, mojolearn
>>> w = np.zeros((64, 8), dtype=np.float32)
>>> opt = mojolearn.RMSprop([w], lr=1e-2)
>>> opt = opt.step([np.ones_like(w)])

__init__(params, lr=0.01, alpha=0.99, eps=1e-08, weight_decay=0.0, momentum=0.0, centered=False, numeric_mode=None)

Methods

load_state_dict(sd)
state_dict()
step(grads)	One update of every param from grads (same shapes, same order).

Attributes

resident_	True once the state lives on the device.
square_avg	Running average of squared gradients, flat over all parameters.
state

property square_avg

Running average of squared gradients, flat over all parameters.