具有多个值的 Tensor 的 bool 值是不明确的 Pytorch 错误

Question

我有一个想要训练的神经网络，但我一直收到 "bool value of Tensor with more than one value is ambiguous" 错误。

这是我的网络：

from torch.autograd import Variable
from torch.utils.data import Dataset, DataLoader
import torch.nn as nn
from torch.nn import functional as F
from DataSetLoader import ReplaysDataSet


class Model(nn.Module):
    def __init__(self):
        super(Model, self).__init__()
        self.l1 = torch.nn.Linear(12,24)
        self.l2 = torch.nn.Linear(24, 20)
        self.l3 = torch.nn.Linear(20, 16)
        self.l4 = torch.nn.Linear(16, 10)
        self.l5 = torch.nn.Linear(10, 6)
        self.l6 = torch.nn.Linear(6, 1)

    def forward(self, t):
        t = F.relu(self.l1(t))
        t = F.relu(self.l2(t))
        t = F.relu(self.l3(t))
        t = F.relu(self.l4(t))
        t = F.relu(self.l5(t))
        y_pred = F.relu(self.l6(t))

        return y_pred
    def get_num_correct(self,preds,labels):
        return preds.argmax(dim=1).eq(labels).sum()

class Test():
    model = Model()
    optimiser = torch.optim.SGD(model.parameters(), lr=0.01)

    dataset = ReplaysDataSet()
    trainLoader = DataLoader(dataset=dataset, batch_size=250, shuffle=True)
    batch = next(iter(trainLoader))
    criterion = nn.MSELoss
    for epoch in range(10):
        totalLoss = 0
        totalCorrect = 0
        for batch in trainLoader:

            data, label = batch
            prediction = model(data)
            prediction = prediction.reshape(250)
            print(prediction)
            print(label)
            loss = criterion(prediction, label)
            optimiser.zero_grad()

            loss.backward()
            optimiser.step()

            totalLoss += loss.item()
            print(totalLoss)
            totalCorrect += model.get_num_correct(prediction, label)
            print(totalCorrect)

这是我的数据加载器

import torch
from torch.autograd import Variable
from torch.utils.data import Dataset, DataLoader
from torch.nn import functional as F

class ReplaysDataSet(Dataset):
    def __init__(self):
        self.xy = np.genfromtxt("dataset.csv", delimiter=',', dtype=np.float32)
        self.length = len(self.xy)
        self.x_data = torch.from_numpy(self.xy[0:,1:])
        self.y_data = torch.from_numpy(self.xy[0:,0])
        self.length = len(self.xy)
    def __getitem__(self, index):
        return self.x_data[index], self.y_data[index]
    def __len__(self):
        return self.length

这里是来自

的 csv im 训练的一些数据

1,303,497,784,748,743,169,462,479,785,310,26,701
1,658,598,645,786,248,381,80,428,248,530,591,145
0,38,490,796,637,130,380,226,359,720,392,464,497
0,94,752,645,801,381,479,475,381,227,645,445,248
0,59,806,254,521,91,538,212,645,609,227,545,531
1,65,254,685,565,91,796,445,658,465,485,472,184
1,385,248,211,612,82,38,485,652,212,373,563,26
1,796,596,785,310,145,479,142,685,748,635,798,474
1,380,658,485,645,36,598,806,428,786,798,645,113
0,743,214,625,642,530,784,645,641,65,598,786,637

我得到的错误是

Traceback (most recent call last):
  File "C:/Users/tayya/PycharmProjects/untitled/NetworkFile.py", line 32, in <module>
    class Test():
  File "C:/Users/tayya/PycharmProjects/untitled/NetworkFile.py", line 50, in Test
    loss = criterion(prediction, label)
File "C:\Users\tayya\PycharmProjects\untitled\venv\lib\site-packages\torch\nn\modules\loss.py", line 428, in __init__
super(MSELoss, self).__init__(size_average, reduce, reduction)
  File "C:\Users\tayya\PycharmProjects\untitled\venv\lib\site-packages\torch\nn\modules\loss.py", line 12, in __init__
    self.reduction = _Reduction.legacy_get_string(size_average, reduce)
  File "C:\Users\tayya\PycharmProjects\untitled\venv\lib\site-packages\torch\nn\_reduction.py", line 36, in legacy_get_string
    if size_average and reduce:
RuntimeError: bool value of Tensor with more than one value is ambiguous

如有任何帮助，我们将不胜感激。我是 NN 的新手，所以如果我犯了一个明显的错误，我深表歉意。

Answer 1

几件事：
1. 根据你的数据，我得出结论，这是一个分类问题，而不是回归问题。因此MSELoss的使用不是很理想，我把它改成BCELoss应该更合适。
2. 你网络的最后一个激活是RELU，因为这是一个二分类问题，所以Sigmoid是更好的选择。
3. 对 "get_num_correct" 函数做了一个小修正。

希望这对你有用：

from torch.utils.data import Dataset, DataLoader
import torch.nn as nn
from torch.nn import functional as F
from DataSetLoader import ReplaysDataSet
import torch

class Model(nn.Module):
    def __init__(self):
        super(Model, self).__init__()
        self.l1 = torch.nn.Linear(12,24)
        self.l2 = torch.nn.Linear(24, 20)
        self.l3 = torch.nn.Linear(20, 16)
        self.l4 = torch.nn.Linear(16, 10)
        self.l5 = torch.nn.Linear(10, 6)
        self.l6 = torch.nn.Linear(6, 1)

    def forward(self, t):
        t = F.relu(self.l1(t))
        t = F.relu(self.l2(t))
        t = F.relu(self.l3(t))
        t = F.relu(self.l4(t))
        t = F.relu(self.l5(t))
        y_pred = F.sigmoid(self.l6(t))

        return y_pred
    def get_num_correct(self,preds,labels):
        return preds.round().squeeze().eq(labels).numpy().sum()

class Test():
    model = Model()
    optimiser = torch.optim.SGD(model.parameters(), lr=0.01)

    dataset = ReplaysDataSet()
    trainLoader = DataLoader(dataset=dataset, batch_size=250, shuffle=True)
    batch = next(iter(trainLoader))
    criterion = nn.BCELoss()
    for epoch in range(10):
        totalLoss = 0
        totalCorrect = 0
        for batch in trainLoader:

            data, label = batch
            prediction = model(data)
            print(prediction)
            print(label)
            loss = criterion(prediction.squeeze(), label)
            optimiser.zero_grad()

            loss.backward()
            optimiser.step()

            totalLoss += loss.item()
            print(totalLoss)
            totalCorrect += model.get_num_correct(prediction, label)
            print(totalCorrect)

具有多个值的 Tensor 的 bool 值是不明确的 Pytorch 错误

bool value of Tensor with more than one value is ambiguous Pytorch error

pytorch

tensor