multilabel classifier
Multi-label categorization tasks are different from multi-categorization tasks, where a multi-categorization task is to classify an instance into a certain category, and multi-label categorization tasks are to classify an instance into more than one category. Multi-label categorization tasks have have two main features:
- The number of class labels is uncertain; some samples may have only one class label, while others may have dozens or even hundreds of class labels.
- The class labels are interdependent, e.g., a sample containing a blue sky class label has a high probability of containing white clouds
As shown in the figure below, which is an example of multi-label classification learning, there are multiple categories in an image, house, tree, cloud, etc., and the deep learning model needs to classify and recognize them one by one.
Multi-label classifier loss function
code implementation
A simplified code implementation of pytorch, a multi-label classifier for images, is shown below.
from torchvision import datasets, transforms
from import DataLoader, Dataset
import torch
import as nn
import as optim
import as F
import os
class CNN():
def __init__(self):
super().__init__()
self.Sq1 = (
nn.Conv2d(in_channels=1, out_channels=16, kernel_size=5, stride=1, padding=2), # (16, 28, 28) # output: (16, 28, 28)
(),
nn.MaxPool2d(kernel_size=2), # (16, 14, 14)
)
self.Sq2 = (
nn.Conv2d(in_channels=16, out_channels=32, kernel_size=5, stride=1, padding=2), # (32, 14, 14)
(),
nn.MaxPool2d(2), # (32, 7, 7)
)
= (32 * 7 * 7, 100)
def forward(self, x):
x = self.Sq1(x)
x = self.Sq2(x)
x = ((0), -1)
x = (x)
## Sigmoid activation
output = (x) # 1/(1+e**(-x))
return output
def loss_fn(pred, target):
return -(target * (pred) + (1 - target) * (1 - pred)).sum()
def multilabel_generate(label):
Y1 = F.one_hot(label, num_classes = 100)
Y2 = F.one_hot(label+10, num_classes = 100)
Y3 = F.one_hot(label+50, num_classes = 100)
multilabel = Y1+Y2+Y3
return multilabel
# def multilabel_generate(label):
# multilabel_dict = {}
# multi_list = []
# for i in range([0]):
# multi_list.append(multilabel_dict[label[i].item()])
# multilabel_tensor = (multi_list)
# return multilabel
def train():
epoches = 10
mnist_net = CNN()
mnist_net.train()
opitimizer = (mnist_net.parameters(), lr=0.002)
mnist_train = ("mnist-data", train=True, download=True, transform=())
train_loader = (mnist_train, batch_size= 128, shuffle=True)
for epoch in range(epoches):
loss = 0
for batch_X, batch_Y in train_loader:
opitimizer.zero_grad()
outputs = mnist_net(batch_X)
loss = loss_fn(outputs, multilabel_generate(batch_Y)) / batch_X.shape[0]
()
()
print(loss)
if __name__ == '__main__':
train()
Above is the python deep learning of multi-label classifier and pytorch source code in detail, more information about multi-label classifier pytorch source code please pay attention to my other related articles!