pytorch loss總結與測試

 

pytorch loss

參考文獻：

https://blog.csdn.net/zhangxb35/article/details/72464152?utm_source=itdadao&utm_medium=referral

loss 測試

import torch
from torch.autograd import Variable

'''
參考文獻：　https://blog.csdn.net/zhangxb35/article/details/72464152?utm_source=itdadao&utm_medium=referral
如果 reduce = False，那么 size_average 參數失效，直接返回向量形式的 loss；
如果 reduce = True，那么 loss 返回的是標量

    如果 size_average = True，返回 loss.mean();
    如果 size_average = True，返回 loss.sum();
'''

# nn.L1Loss: loss(input, target)=|input-target|
if False:
    loss_fn = torch.nn.L1Loss(reduce=True, size_average=False)
    input = torch.autograd.Variable(torch.randn(3, 4))
    target = torch.autograd.Variable(torch.randn(3, 4))
    loss = loss_fn(input, target)
    print(input)
    print(target)
    print(loss)
    print(input.size(), target.size(), loss.size())


# nn.SmoothL1Loss 　在(-1, 1)上是平方loss, 其他情況是L1 loss
if False:
    loss_fn = torch.nn.SmoothL1Loss(reduce=False, size_average=False)
    input = torch.autograd.Variable(torch.randn(3, 4))
    target = torch.autograd.Variable(torch.randn(3, 4))
    loss = loss_fn(input, target)
    print(input)
    print(target)
    print(loss)
    print(input.size(), target.size(), loss.size())

# nn.MSELoss  均方損失函數
if False:
    loss_fn = torch.nn.MSELoss(reduce=False, size_average=False)
    input = torch.autograd.Variable(torch.randn(3, 4))
    target = torch.autograd.Variable(torch.randn(3, 4))
    loss = loss_fn(input, target)
    print(input)
    print(target)
    print(loss)
    print(input.size(), target.size(), loss.size())

# nn.BCELoss
if False:
    import torch.nn.functional as F

    loss_fn = torch.nn.BCELoss(reduce=False, size_average=False)
    input = torch.autograd.Variable(torch.randn(3, 4))
    target = torch.autograd.Variable(torch.FloatTensor(3, 4).random_(2))
    loss = loss_fn(F.sigmoid(input), target)
    print(input, input.shape)
    print(F.sigmoid(input))
    print(target, target.shape)
    print(loss, loss.shape)

# nn.CrossEntropyLoss
if False:
    weight = torch.Tensor([1, 2, 1, 1, 10])
    loss_fn = torch.nn.CrossEntropyLoss(reduce=False, size_average=False, weight=None)
    input = Variable(torch.randn(3, 5))  # (batch_size, C)
    target = Variable(torch.LongTensor(3).random_(5))
    loss = loss_fn(input, target)
    print(input)
    print(target)
    print(loss)
# nn.NLLLoss 負的　log likehood loss損失.用於訓練一個ｎ類分類器.
if False:
    m = nn.LogSoftmax()
    loss = nn.NLLLoss()
    # input is of size nBatch x nClasses = 3 x 5
    input = Variable(torch.randn(3, 5), requires_grad=True)
    # each element in target has to have 0 <= value < nclasses
    target = Variable(torch.LongTensor([1, 0, 4]))
    output = loss(m(input), target)
    print(output)

# nn.LLLoss2d  對於圖片的　negtive log likehood loss.計算每個像素的NLL loss
if False:
    m = nn.Conv2d(16, 32, (3, 3)).float()
    loss = nn.NLLLoss2d()
    # input is of size nBatch x nClasses x height x width
    input = Variable(torch.randn(3, 16, 10, 10))
    # each element in target has to have 0 <= value < nclasses
    target = Variable(torch.LongTensor(3, 8, 8).random_(0, 4))
    output = loss(m(input), target)
    print(m(input))
    print(output)

# nn.MultiLabelMarginLoss  input : x --> (N, c), y --> (N, c)其中y是 LongTensor, 且其元素為類別的index
if False:
    x=Variable(torch.randn(3, 4))
    y=Variable(torch.LongTensor(3, 4).random_(4))
    loss=torch.nn.MultiLabelMarginLoss()
    output = loss(x, y)
    print(output)

# nn.MultiLabelSoftMarginLoss  與MultiLableMarginLoss相同，區別在於y的類型是FloatTensor
if False:
    x = Variable(torch.randn(3, 10))
    y = Variable(torch.FloatTensor(3, 10).random_(10))
    loss = torch.nn.MultiLabelSoftMarginLoss()
    output = loss(x, y)
    print(output)

# nn.MultiMarginLoss 適用於多分類模型
if True:
    x = Variable(torch.randn(3, 10))
    y = Variable(torch.LongTensor(3).random_(10))
    loss = torch.nn.MultiMarginLoss()
    output=loss(x, y)
    print(output)