#!/usr/bin/env python3
import torch
from ..utils.broadcasting import _mul_broadcast_shape
from .mean import Mean
[docs]class ZeroMean(Mean):
def __init__(self, batch_shape=torch.Size(), **kwargs):
super(ZeroMean, self).__init__()
self.batch_shape = batch_shape
def forward(self, input):
mean = torch.zeros(*self.batch_shape, 1, dtype=input.dtype, device=input.device)
if input.shape[:-2] == self.batch_shape:
return mean.expand(input.shape[:-1])
else:
return mean.expand(_mul_broadcast_shape(input.shape[:-1], mean.shape))