Differences with torch.nn.GaussianNLLLoss

torch.nn.GaussianNLLLoss

class torch.nn.GaussianNLLLoss(
    *,
    full=False,
    eps=1e-06,
    reduction='mean'
)(input, target, var) -> Tensor/Scalar

For more information, see torch.nn.GaussianNLLLoss.

mindspore.nn.GaussianNLLLoss

class mindspore.nn.GaussianNLLLoss(
    *,
    full=False,
    eps=1e-06,
    reduction='mean'
)(logits, labels, var) -> Tensor/Scalar

For more information, see mindspore.nn.GaussianNLLLoss.

Differences

PyTorch: Obey the negative log-likelihood loss of the Gaussian distribution.

MindSpore: Implements the same function as PyTorch. If there is a number less than 0 in var, PyTorch will report an error directly, while MindSpore will calculate max(var, eps) and then pass the result to log for calculation.

Categories	Subcategories	PyTorch	MindSpore	Differences
Parameters	Parameter 1	full	full	Same function
	Parameter 2	eps	eps	Same function
	Parameter 3	reduction	reduction	Same function
Inputs	Input 1	input	logits	Same function, different parameter names
	Input 2	target	labels	Same function, different parameter names
	Input 3	var	var	Same function

Code Example

The two APIs implement basically the same functionality and usage, but PyTorch and MindSpore handle the case of input var<0 differently.

# PyTorch
import torch
from torch import nn
import numpy as np

arr1 = np.arange(8).reshape((4, 2))
arr2 = np.array([2, 3, 1, 4, 6, 4, 4, 9]).reshape((4, 2))
logits = torch.tensor(arr1, dtype=torch.float32)
labels = torch.tensor(arr2, dtype=torch.float32)
loss = nn.GaussianNLLLoss(reduction='mean')
var = torch.tensor(np.ones((4, 1)), dtype=torch.float32)
output = loss(logits, labels, var)
# tensor(1.4375)

# If there are elements in the var that are less than 0, PyTorch will directly report an error
var[0] = -1
output2 = loss(logits, labels, var)
# ValueError: var has negative entry/entries

# MindSpore
import numpy as np
from mindspore import Tensor
import mindspore.nn as nn
from mindspore import dtype as mstype

arr1 = np.arange(8).reshape((4, 2))
arr2 = np.array([2, 3, 1, 4, 6, 4, 4, 9]).reshape((4, 2))
logits = Tensor(arr1, mstype.float32)
labels = Tensor(arr2, mstype.float32)
loss = nn.GaussianNLLLoss(reduction='mean')
var = Tensor(np.ones((4, 1)), mstype.float32)
output = loss(logits, labels, var)
print(output)
# 1.4374993

# If there are elements that are less than 0 in var, MindSpore will use the result of max(var, eps)
var[0] = -1
output2 = loss(logits, labels, var)
print(output2)
# 499999.22