Pytorch学习记录-Pytorch可视化使用tensorboardX
在很早很早以前(至少一个半月),我做过几节关于tensorboard的学习记录。
https://www.jianshu.com/p/23205a7921cd
https://www.jianshu.com/p/6235c1ecde67
https://www.jianshu.com/p/2b24454b0629
https://www.jianshu.com/p/0080047e5456
迟迟没有转到Pytorch的原因也是tensorflow的可视化做的好,不过现在Pytorch也支持了,在教程里有,学习一个。
在本教程中,使用简单的神经网络实现MNIST分类器,并使用TensorBoard可视化训练过程。在训练阶段,我们通过scalar_summary绘制损失和准确度函数,并通过image_summary可视化训练图像。此外,我们使用histogram_summary可视化神经网络参数的权重和梯度值。
pytorch使用tensorboard有三种方法:
昨天看了一下余霆嵩的教程,推荐使用tensorboardX,使用比logger更方便一些。
注意看注释就行了,这里我没有生成更复杂的直方图,仅仅记录了Loss、Accuracy、Graph
import torchimport torch.nn as nnimport torchvisionfrom torchvision import transformsfrom logger import Loggerfrom tensorboardX import SummaryWriter 加载SummaryWriter,设置保存地址。writer = SummaryWriter('./logs')# Device configurationdevice = torch.device('cuda' if torch.cuda.is_available() else 'cpu') MNIST datasetdataset = torchvision.datasets.MNIST(root='./data', train=True, transform=transforms.ToTensor(), download=True)#Data loaderdata_loader = torch.utils.data.DataLoader(dataset=dataset, batch_size=100, shuffle=True)# lly connected neural network with one hidden layerclass NeuralNet(nn.Module): def __init__(self, input_size=784, hidden_size=500, num_classes=10): super(NeuralNet, self).__init__() self.fc1 = nn.Linear(input_size, hidden_size) self.relu = nn.ReLU() self.fc2 = nn.Linear(hidden_size, num_classes) df forward(self, x): out = self.fc1(x) out = self.relu(out) out = self.fc2(out) return outmodel NeuralNet().to(device)# 损失函数和化器criterion = nn.CrossEntropyLoss()optimizer = torch.optim.Adam(model.parameters(), lr=0.00001)data_iter= iter(data_loader)iter_per_epoch = len(data_loader)total_step = 10000# Start tainingfor step in range(total_step): # Resetthe data_iter if (step + 1) % iter_per_epoch == 0: data_iter = iter(data_loader) # Fetchimages and labels images, labels = next(data_iter) # view作用是将多行tensor拼接为一行,reshape张量形状,如果你不知道你想要多少行,但确定列数,那么你可以将行数设置为-1(同样,不知道多少列,可以将列数设为-1) # size获取images的信息(行数,列数) images, labels = images.view(images.size(0), -1).to(device), labels.to(device) writer.add_graph(model, (images,)) # Forward pass outputs = model(images) loss = criterion(outputs, labels) writer.add_scalar('Loss', loss, step + 1) # Backwar and optimize optimizer.zero_grad() loss.backward() optimizer.step() # Computeaccuracy _, argmax = torch.max(outputs, 1) accuracy = (labels == argmax.squeeze()).float().mean() writer.add_scalar('accuracy', accuracy, step + 1) if (step + ) % 100 == 0: print('Step [{}/{}], Loss: {:.4f}, Acc: {:.2f}'.format(step + 1, total_step, loss.item(), accuracy.item()))
搞定之后会在文件列表里看到一个logs文件夹,记录就在里面。
在根目录下命令行输入
tensorboard --logidr logs
得到反馈后在浏览器输入"http://localhost:6006"就可以进入tensorboard。