utils package

Submodules

utils.device module

utils.device.get_device(tensor)

Get the device of a tensor

Parameters:

tensor: Tensor

tensor from which to extract the device

Returns:

computation device (0-n or cpu)

utils.device.to_same_device(*tensors, device=None)

Move all tensors to the same device (by default the first tensor’s one)

Parameters:

tensors: Tensor

tensors which we want to move to a common device

Returns:

a tuple of tensors in the same order as the arguments moved to a common device

utils.helpers module

utils.helpers.batchify(func, batch, samples=[], **args)

utils.helpers.get_token_first_indices(x, token)

utils.moving_average module

class utils.moving_average.MovingAverage(init_value=0)

Bases: object

Keeps a moving average of some quantity

add(new_value, new_weight=1)

Includes a single value into a moving average

Parameters:

new_value: float

pointwise estimate of the quantity to estimate

new_weight: float

a weight by which to ponder the new value

add_array(new_values)

Includes an array of values into the moving average

Parameters:

new_values: torch/np array of floats

pointwise estimates of the quantity to estimate

reset()

Resets the moving average to its initial conditions

utils.moving_average.average(moving_averages)

utils.observable module

class utils.observable.Observable

Bases: object

Textbook observer pattern

dispatch(*args, **kwargs)

enroll(function)

utils.observable.forward(observable1, observable2)

Forwards messages from an observable to another with the same signature

Module contents