mighty.utils.var_online.VarianceOnlineLabels

class mighty.utils.var_online.VarianceOnlineLabels

Keep track of population mean and std for each unique class label.

Methods

__init__()
activate(is_active)	Activates or deactivates the updates.
get_mean()	Returns:
get_mean_labels()	Returns:
get_mean_std([unbiased])	Return the mean and std for each unique label individually without the labels themselves.
get_mean_std_labels([unbiased])	Return the mean and std for each unique label individually.
labels()	Returns:
reset()	Reset the mean and the count.
update(tensor, labels)	Update sample mean (and variance) from a batch of new values, split by labels.

activate(is_active: bool)

Activates or deactivates the updates.

Parameters:

is_activebool

New state.

get_mean()

Returns:

mean_sorted(C, V) torch.Tensor

Mean tensor for each of C unique class labels.

get_mean_labels()

Returns:

mean_sorted(C, V) torch.Tensor

Mean tensor for each of C unique class labels.

labels_sorted(C,) torch.Tensor

Class labels, associated with mean_sorted.

get_mean_std(unbiased=True)

Return the mean and std for each unique label individually without the labels themselves.

Parameters:

unbiasedbool, optional

Biased (False) or unbiased (True) variance estimate. Default: True

Returns:

mean_sorted(C, V) torch.Tensor

Mean tensor for each of C unique class labels.

std_sorted(C, V) torch.Tensor

Std tensor for each of C unique class labels.

get_mean_std_labels(unbiased=True)

Return the mean and std for each unique label individually.

Parameters:

unbiasedbool, optional

Biased (False) or unbiased (True) variance estimate. Default: True

Returns:

mean_sorted(C, V) torch.Tensor

Mean tensor for each of C unique class labels.

std_sorted(C, V) torch.Tensor

Std tensor for each of C unique class labels.

labels_sorted(C,) torch.Tensor

Class labels, associated with mean_sorted and std_sorted.

labels()

Returns:

list

Unique sorted class labels.

reset()

Reset the mean and the count.

update(tensor, labels)

Update sample mean (and variance) from a batch of new values, split by labels.

Parameters:

tensor(B, V) torch.Tensor

A tensor sample.

labels(B,) torch.Tensor

Batch labels.