Differences with torch.nn.TripletMarginLoss

The following mapping relationships can be found in this file.

PyTorch APIs	MindSpore APIs
torch.nn.TripletMarginLoss	mindspore.nn.TripletMarginLoss
torch.functional.triplet_margin_loss	mindspore.ops.triplet_margin_loss

torch.nn.TripletMarginLoss

torch.nn.TripletMarginLoss(
    margin=1.0,
    p=2.0,
    eps=1e-06,
    swap=False,
    size_average=None,
    reduce=None,
    reduction='mean'
)(anchor, positive, negative) -> Tensor/Scalar

For more information, see torch.nn.TripletMarginLoss.

mindspore.nn.TripletMarginLoss

mindspore.nn.TripletMarginLoss(
    p=2,
    swap=False,
    eps=1e-06,
    reduction='mean'
)(margin, x, positive, negative) -> Tensor/Scalar

For more information, see mindspore.nn.TripletMarginLoss.

Differences

API function of MindSpore is consistent with that of PyTorch.

PyTorch:

• PyTorch has two initialization parameters size_average and reduce , which are deprecated and replaced by reduction .

• margin is an initialization parameter, not an input parameter. The data type of margin is float.

MindSpore:

• MindSpore doesn’t have initialization parameters size_average and reduce .

• margin is not an initialization parameter, but an input parameter. The data type of margin can be tensor or float.

• The input parameter x of MindSpore corresponds to the input parameter anchor of PyTorch.

• MindSpore’s initialization parameters ‘swap’ and ‘eps’ are positioned in a different order than PyTorch.

Categories	Subcategories	PyTorch	MindSpore	Differences
Parameters	Parameter 1	margin	-	Different position, same function. Data type is float.
	Parameter 2	p	p	-
	Parameter 3	eps	swap	Different position, same function.
	Parameter 4	swap	eps	Different position, same function.
	Parameter 5	size_average	-	PyTorch has deprecated this parameter, while MindSpore does not have this parameter.
	Parameter 6	reduce	-	PyTorch has deprecated this parameter, while MindSpore does not have this parameter.
	Parameter 7	reduction	reduction	-
Input	Input 1	-	margin	Different position, same function. Data type can be tensor or float.
	Input 2	anchor	x	Same function, different parameter names.
	Input 3	positive	positive	-
	Input 4	negative	negative	-

Code Example

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

p = 2
swap = False
eps = 1e-06
reduction = 'mean'
margin = 1.0
triplet_margin_loss = nn.TripletMarginLoss(margin=margin, p=p, eps=eps, swap=swap, reduction=reduction)

x = torch.tensor(np.array([[0.3, 0.7], [0.5, 0.5]]), dtype=torch.float32)
positive = torch.tensor(np.array([[0.4, 0.6], [0.4, 0.6]]), dtype=torch.float32)
negative = torch.tensor(np.array([[0.2, 0.9], [0.3, 0.7]]), dtype=torch.float32)
output = triplet_margin_loss(x, positive, negative)
print(output)
# tensor(0.8882)

# MindSpore
import mindspore as ms
import mindspore.nn as nn
import numpy as np

p = 2
swap = False
eps = 1e-06
reduction = 'mean'
triplet_margin_loss = nn.TripletMarginLoss(p=p, swap=swap, eps=eps, reduction=reduction)

x = ms.Tensor(np.array([[0.3, 0.7], [0.5, 0.5]]), dtype=ms.float32)
positive = ms.Tensor(np.array([[0.4, 0.6], [0.4, 0.6]]), dtype=ms.float32)
negative = ms.Tensor(np.array([[0.2, 0.9], [0.3, 0.7]]), dtype=ms.float32)
margin = ms.Tensor(1.0, ms.float32)
output = triplet_margin_loss(x, positive, negative, margin)
print(output)
# 0.8881968