535 KiB
%matplotlib inlineTraining a Classifier
This is it. You have seen how to define neural networks, compute loss and make updates to the weights of the network.
Now you might be thinking,
What about data?
Generally, when you have to deal with image, text, audio or video data,
you can use standard python packages that load data into a numpy array.
Then you can convert this array into a torch.*Tensor.
- For images, packages such as Pillow, OpenCV are useful
- For audio, packages such as scipy and librosa
- For text, either raw Python or Cython based loading, or NLTK and SpaCy are useful
Specifically for vision, we have created a package called
torchvision, that has data loaders for common datasets such as
Imagenet, CIFAR10, MNIST, etc. and data transformers for images, viz.,
torchvision.datasets and torch.utils.data.DataLoader.
This provides a huge convenience and avoids writing boilerplate code.
For this tutorial, we will use the CIFAR10 dataset. It has the classes: ‘airplane’, ‘automobile’, ‘bird’, ‘cat’, ‘deer’, ‘dog’, ‘frog’, ‘horse’, ‘ship’, ‘truck’. The images in CIFAR-10 are of size 3x32x32, i.e. 3-channel color images of 32x32 pixels in size.
Training an image classifier
We will do the following steps in order:
- Load and normalizing the CIFAR10 training and test datasets using
torchvision - Define a Convolutional Neural Network
- Define a loss function
- Train the network on the training data
- Test the network on the test data
Loading and normalizing CIFAR10
Using torchvision, it’s extremely easy to load CIFAR10. The output of torchvision datasets are PILImage images of range [0, 1]. We transform them to Tensors of normalized range [-1, 1].
import torch
import torchvision
import torchvision.transforms as transforms
transform = transforms.Compose(
[transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
trainset = torchvision.datasets.CIFAR10(root='./data', train=True,
download=True, transform=transform)
trainloader = torch.utils.data.DataLoader(trainset, batch_size=4,
shuffle=True, num_workers=2)
testset = torchvision.datasets.CIFAR10(root='./data', train=False,
download=True, transform=transform)
testloader = torch.utils.data.DataLoader(testset, batch_size=4,
shuffle=False, num_workers=2)
classes = ('plane', 'car', 'bird', 'cat',
'deer', 'dog', 'frog', 'horse', 'ship', 'truck')Files already downloaded and verified Files already downloaded and verified
Let us show some of the training images, for fun.
import matplotlib.pyplot as plt
import numpy as np
# functions to show an image
def imshow(img):
img = img / 2 + 0.5 # unnormalize
npimg = img.numpy()
plt.imshow(np.transpose(npimg, (1, 2, 0)))
plt.show()
# get some random training images
dataiter = iter(trainloader)
images, labels = dataiter.next()
# show images
imshow(torchvision.utils.make_grid(images))
# print labels
print(' '.join('%5s' % classes[labels[j]] for j in range(4)))car deer deer dog
Define a Convolutional Neural Network
Copy the neural network from the Neural Networks section before and modify it to take 3-channel images (instead of 1-channel images as it was defined).
import torch.nn as nn
import torch.nn.functional as F
class Net(nn.Module):
def __init__(self):
super(Net, self).__init__()
self.conv1 = nn.Conv2d(3, 6, 5)
self.pool = nn.MaxPool2d(2, 2)
self.conv2 = nn.Conv2d(6, 16, 5)
self.fc1 = nn.Linear(16 * 5 * 5, 120)
self.fc2 = nn.Linear(120, 84)
self.fc3 = nn.Linear(84, 10)
def forward(self, x):
x = self.pool(F.relu(self.conv1(x)))
x = self.pool(F.relu(self.conv2(x)))
x = x.view(-1, 16 * 5 * 5)
x = F.relu(self.fc1(x))
x = F.relu(self.fc2(x))
x = self.fc3(x)
return x
net = Net()Define a Loss function and optimizer
Let's use a Classification Cross-Entropy loss and SGD with momentum.
import torch.optim as optim
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(net.parameters(), lr=0.001, momentum=0.9)Train the network
This is when things start to get interesting. We simply have to loop over our data iterator, and feed the inputs to the network and optimize.
%%time
for epoch in range(2): # loop over the dataset multiple times
running_loss = 0.0
for i, data in enumerate(trainloader, 0):
# get the inputs; data is a list of [inputs, labels]
inputs, labels = data
# zero the parameter gradients
optimizer.zero_grad()
# forward + backward + optimize
outputs = net(inputs)
loss = criterion(outputs, labels)
loss.backward()
optimizer.step()
# print statistics
running_loss += loss.item()
if i % 2000 == 1999: # print every 2000 mini-batches
print('[%d, %5d] loss: %.3f' %
(epoch + 1, i + 1, running_loss / 2000))
running_loss = 0.0
print('Finished Training')[1, 2000] loss: 2.205 [1, 4000] loss: 1.840 [1, 6000] loss: 1.665 [1, 8000] loss: 1.586 [1, 10000] loss: 1.511 [1, 12000] loss: 1.460 [2, 2000] loss: 1.376 [2, 4000] loss: 1.371 [2, 6000] loss: 1.360 [2, 8000] loss: 1.303 [2, 10000] loss: 1.281 [2, 12000] loss: 1.261 Finished Training CPU times: user 31min, sys: 32.8 s, total: 31min 33s Wall time: 2min 19s
Let's quickly save our trained model:
PATH = './cifar_net.pth'
torch.save(net.state_dict(), PATH)See here <https://pytorch.org/docs/stable/notes/serialization.html>_
for more details on saving PyTorch models.
Test the network on the test data
We have trained the network for 2 passes over the training dataset. But we need to check if the network has learnt anything at all.
We will check this by predicting the class label that the neural network outputs, and checking it against the ground-truth. If the prediction is correct, we add the sample to the list of correct predictions.
Okay, first step. Let us display an image from the test set to get familiar.
dataiter = iter(testloader)
images, labels = dataiter.next()
# print images
imshow(torchvision.utils.make_grid(images))
print('GroundTruth: ', ' '.join('%5s' % classes[labels[j]] for j in range(4)))GroundTruth: cat ship ship plane
Next, let's load back in our saved model (note: saving and re-loading the model wasn't necessary here, we only did it to illustrate how to do so):
net = Net()
net.load_state_dict(torch.load(PATH))<All keys matched successfully>
Okay, now let us see what the neural network thinks these examples above are:
outputs = net(images)
print(outputs)tensor([[-1.4235, -1.9955, 0.6154, 2.4929, 0.1378, 1.3219, 0.6264, -0.5958,
-0.6519, -0.7850],
[ 4.3612, 4.4116, -2.2258, -2.9321, -2.9944, -4.7716, -5.0000, -1.8543,
4.5418, 4.4014],
[ 2.2970, 3.3485, -0.9210, -1.8730, -2.9894, -3.0735, -3.1630, -1.9718,
3.5772, 3.1711],
[ 3.6032, 0.0659, 0.9458, -1.2418, -0.3086, -2.9271, -2.4405, -1.6286,
2.8094, 0.2056]], grad_fn=<AddmmBackward>)
The outputs are energies for the 10 classes. The higher the energy for a class, the more the network thinks that the image is of the particular class. So, let's get the index of the highest energy:
_, predicted = torch.max(outputs, 1)
print('Predicted: ', ' '.join('%5s' % classes[predicted[j]]
for j in range(4)))Predicted: cat ship ship plane
The results seem pretty good.
Let us look at how the network performs on the whole dataset.
correct = 0
total = 0
with torch.no_grad():
for data in testloader:
images, labels = data
outputs = net(images)
_, predicted = torch.max(outputs.data, 1)
total += labels.size(0)
correct += (predicted == labels).sum().item()
print('Accuracy of the network on the 10000 test images: %d %%' % (
100 * correct / total))Accuracy of the network on the 10000 test images: 55 %
That looks way better than chance, which is 10% accuracy (randomly picking a class out of 10 classes). Seems like the network learnt something.
Hmmm, what are the classes that performed well, and the classes that did not perform well:
class_correct = list(0. for i in range(10))
class_total = list(0. for i in range(10))
with torch.no_grad():
for data in testloader:
images, labels = data
outputs = net(images)
_, predicted = torch.max(outputs, 1)
c = (predicted == labels).squeeze()
for i in range(4):
label = labels[i]
class_correct[label] += c[i].item()
class_total[label] += 1
for i in range(10):
print('Accuracy of %5s : %2d %%' % (
classes[i], 100 * class_correct[i] / class_total[i]))Accuracy of plane : 55 % Accuracy of car : 52 % Accuracy of bird : 54 % Accuracy of cat : 46 % Accuracy of deer : 51 % Accuracy of dog : 21 % Accuracy of frog : 64 % Accuracy of horse : 62 % Accuracy of ship : 73 % Accuracy of truck : 70 %
Okay, so what next?
How do we run these neural networks on the GPU?
Training on GPU
Just like how you transfer a Tensor onto the GPU, you transfer the neural net onto the GPU.
Let's first define our device as the first visible cuda device if we have CUDA available:
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
# Assuming that we are on a CUDA machine, this should print a CUDA device:
print(device)cuda:0
The rest of this section assumes that device is a CUDA device.
Then these methods will recursively go over all modules and convert their parameters and buffers to CUDA tensors:
Remember that you will have to send the inputs and targets at every step to the GPU too:
%%time
BATCH_SIZE = 1024
EPOCHS = 60
OUTPUTS= 1
LR = 0.025
MINI_BATCH_SIZE = int((50000/BATCH_SIZE)/OUTPUTS)
print(MINI_BATCH_SIZE)
net = Net()
net.to(device)
trainset = torchvision.datasets.CIFAR10(root='./data', train=True,
download=True, transform=transform)
trainloader = torch.utils.data.DataLoader(trainset, batch_size=BATCH_SIZE,
shuffle=True, num_workers=6)
testset = torchvision.datasets.CIFAR10(root='./data', train=False,
download=True, transform=transform)
testloader = torch.utils.data.DataLoader(testset, batch_size=BATCH_SIZE,
shuffle=False, num_workers=6)
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(net.parameters(), lr=LR, momentum=0.9)
for epoch in range(EPOCHS): # loop over the dataset multiple times
running_loss = 0.0
for i, data in enumerate(trainloader, 0):
# get the inputs; data is a list of [inputs, labels]
inputs, labels = data[0].to(device), data[1].to(device)
# zero the parameter gradients
optimizer.zero_grad()
# forward + backward + optimize
outputs = net(inputs)
loss = criterion(outputs, labels)
loss.backward()
optimizer.step()
# print statistics
running_loss += loss.item()
if i % MINI_BATCH_SIZE == MINI_BATCH_SIZE - 1: # print every 2000 mini-batches
print('[%d, %5d] loss: %.3f' %
(epoch + 1, i + 1, running_loss / MINI_BATCH_SIZE))
running_loss = 0.0
correct = 0
total = 0
with torch.no_grad():
for data in testloader:
images, labels = data[0].to(device), data[1].to(device)
outputs = net(images)
_, predicted = torch.max(outputs.data, 1)
total += labels.size(0)
correct += (predicted == labels).sum().item()
print('Accuracy of the network on the 10000 test images: %.2f %%' % (
100.0 * correct / total))
print('Finished Training')48 Files already downloaded and verified Files already downloaded and verified [1, 48] loss: 2.283 Accuracy of the network on the 10000 test images: 18.28 % [2, 48] loss: 2.052 Accuracy of the network on the 10000 test images: 29.48 % [3, 48] loss: 1.807 Accuracy of the network on the 10000 test images: 37.51 % [4, 48] loss: 1.635 Accuracy of the network on the 10000 test images: 42.83 % [5, 48] loss: 1.528 Accuracy of the network on the 10000 test images: 46.74 % [6, 48] loss: 1.448 Accuracy of the network on the 10000 test images: 48.84 % [7, 48] loss: 1.386 Accuracy of the network on the 10000 test images: 51.40 % [8, 48] loss: 1.322 Accuracy of the network on the 10000 test images: 51.35 % [9, 48] loss: 1.286 Accuracy of the network on the 10000 test images: 54.47 % [10, 48] loss: 1.226 Accuracy of the network on the 10000 test images: 55.97 % [11, 48] loss: 1.198 Accuracy of the network on the 10000 test images: 55.41 % [12, 48] loss: 1.169 Accuracy of the network on the 10000 test images: 57.73 % [13, 48] loss: 1.144 Accuracy of the network on the 10000 test images: 58.52 % [14, 48] loss: 1.107 Accuracy of the network on the 10000 test images: 58.70 % [15, 48] loss: 1.084 Accuracy of the network on the 10000 test images: 60.21 % [16, 48] loss: 1.050 Accuracy of the network on the 10000 test images: 60.70 % [17, 48] loss: 1.031 Accuracy of the network on the 10000 test images: 60.42 % [18, 48] loss: 1.002 Accuracy of the network on the 10000 test images: 59.75 % [19, 48] loss: 0.996 Accuracy of the network on the 10000 test images: 62.09 % [20, 48] loss: 0.960 Accuracy of the network on the 10000 test images: 61.42 % [21, 48] loss: 0.938 Accuracy of the network on the 10000 test images: 62.70 % [22, 48] loss: 0.923 Accuracy of the network on the 10000 test images: 62.51 % [23, 48] loss: 0.912 Accuracy of the network on the 10000 test images: 61.14 % [24, 48] loss: 0.899 Accuracy of the network on the 10000 test images: 62.50 % [25, 48] loss: 0.874 Accuracy of the network on the 10000 test images: 63.09 % [26, 48] loss: 0.860 Accuracy of the network on the 10000 test images: 63.33 % [27, 48] loss: 0.838 Accuracy of the network on the 10000 test images: 63.39 % [28, 48] loss: 0.809 Accuracy of the network on the 10000 test images: 63.15 % [29, 48] loss: 0.787 Accuracy of the network on the 10000 test images: 63.53 % [30, 48] loss: 0.781 Accuracy of the network on the 10000 test images: 63.27 % [31, 48] loss: 0.768 Accuracy of the network on the 10000 test images: 62.57 % [32, 48] loss: 0.748 Accuracy of the network on the 10000 test images: 64.45 % [33, 48] loss: 0.717 Accuracy of the network on the 10000 test images: 64.04 % [34, 48] loss: 0.702 Accuracy of the network on the 10000 test images: 62.87 % [35, 48] loss: 0.699 Accuracy of the network on the 10000 test images: 63.34 % [36, 48] loss: 0.681 Accuracy of the network on the 10000 test images: 62.12 % [37, 48] loss: 0.671 Accuracy of the network on the 10000 test images: 63.91 % [38, 48] loss: 0.657 Accuracy of the network on the 10000 test images: 63.17 % [39, 48] loss: 0.639 Accuracy of the network on the 10000 test images: 63.84 % [40, 48] loss: 0.628 Accuracy of the network on the 10000 test images: 63.08 % [41, 48] loss: 0.616 Accuracy of the network on the 10000 test images: 62.92 % [42, 48] loss: 0.614 Accuracy of the network on the 10000 test images: 62.51 % [43, 48] loss: 0.590 Accuracy of the network on the 10000 test images: 62.78 % [44, 48] loss: 0.572 Accuracy of the network on the 10000 test images: 62.72 % [45, 48] loss: 0.576 Accuracy of the network on the 10000 test images: 62.31 % [46, 48] loss: 0.547 Accuracy of the network on the 10000 test images: 62.09 % [47, 48] loss: 0.548 Accuracy of the network on the 10000 test images: 61.66 % [48, 48] loss: 0.554 Accuracy of the network on the 10000 test images: 62.38 % [49, 48] loss: 0.513 Accuracy of the network on the 10000 test images: 62.48 % [50, 48] loss: 0.510 Accuracy of the network on the 10000 test images: 62.18 % [51, 48] loss: 0.505 Accuracy of the network on the 10000 test images: 60.70 % [52, 48] loss: 0.487 Accuracy of the network on the 10000 test images: 61.94 % [53, 48] loss: 0.472 Accuracy of the network on the 10000 test images: 62.13 % [54, 48] loss: 0.468 Accuracy of the network on the 10000 test images: 62.50 % [55, 48] loss: 0.438 Accuracy of the network on the 10000 test images: 61.41 % [56, 48] loss: 0.442 Accuracy of the network on the 10000 test images: 61.57 % [57, 48] loss: 0.416 Accuracy of the network on the 10000 test images: 60.88 % [58, 48] loss: 0.419 Accuracy of the network on the 10000 test images: 61.49 % [59, 48] loss: 0.425 Accuracy of the network on the 10000 test images: 60.62 % [60, 48] loss: 0.412 Accuracy of the network on the 10000 test images: 61.49 % Finished Training CPU times: user 32.8 s, sys: 26 s, total: 58.8 s Wall time: 2min 37s
Why dont I notice MASSIVE speedup compared to CPU? Because your network is really small.
Exercise: Try increasing the width of your network (argument 2 of
the first nn.Conv2d, and argument 1 of the second nn.Conv2d –
they need to be the same number), see what kind of speedup you get.
Goals achieved:
- Understanding PyTorch's Tensor library and neural networks at a high level.
- Train a small neural network to classify images
Training on multiple GPUs
If you want to see even more MASSIVE speedup using all of your GPUs,
please check out :doc:data_parallel_tutorial.
Where do I go next?
- :doc:
Train neural nets to play video games </intermediate/reinforcement_q_learning> -
Train a state-of-the-art ResNet network on imagenet_ -
Train a face generator using Generative Adversarial Networks_ -
Train a word-level language model using Recurrent LSTM networks_ -
More examples_ -
More tutorials_ -
Discuss PyTorch on the Forums_ -
Chat with other users on Slack_