Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,148 @@
|
|
| 1 |
---
|
| 2 |
license: mit
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: mit
|
| 3 |
---
|
| 4 |
+
Libraries needed:
|
| 5 |
+
```
|
| 6 |
+
import torch
|
| 7 |
+
import torchvision
|
| 8 |
+
import torchvision.transforms as transforms
|
| 9 |
+
from tqdm import tqdm
|
| 10 |
+
from torch import nn
|
| 11 |
+
import matplotlib.pyplot as plt
|
| 12 |
+
```
|
| 13 |
+
|
| 14 |
+
|
| 15 |
+
|
| 16 |
+
|
| 17 |
+
to define a data loader
|
| 18 |
+
```
|
| 19 |
+
transformRes = transforms.Compose([
|
| 20 |
+
transforms.ToTensor(),
|
| 21 |
+
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
|
| 22 |
+
# transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
|
| 23 |
+
])
|
| 24 |
+
|
| 25 |
+
trainsetRes = torchvision.datasets.CIFAR10(root='./data', train=True, download=True, transform=transformRes)
|
| 26 |
+
trainloaderRes64 = torch.utils.data.DataLoader(trainsetRes, batch_size=64, shuffle=True, num_workers=10)
|
| 27 |
+
|
| 28 |
+
testsetRes = torchvision.datasets.CIFAR10(root='./data', train=False, download=True, transform=transformRes)
|
| 29 |
+
testloaderRes64 = torch.utils.data.DataLoader(testsetRes, batch_size=64, shuffle=False, num_workers=10)
|
| 30 |
+
```
|
| 31 |
+
|
| 32 |
+
|
| 33 |
+
The model itself and training
|
| 34 |
+
```
|
| 35 |
+
import torchvision.models as models
|
| 36 |
+
|
| 37 |
+
# Load the pretrained model from pytorch
|
| 38 |
+
resnet50v2 = models.resnet50(pretrained=True)
|
| 39 |
+
|
| 40 |
+
# Freeze the parameters of the model
|
| 41 |
+
for param in resnet50v2.parameters():
|
| 42 |
+
param.requires_grad = True
|
| 43 |
+
|
| 44 |
+
|
| 45 |
+
# Change the final layer to match the number of classes in the CIFAR-10 dataset
|
| 46 |
+
num_ftrs = resnet50v2.fc.in_features
|
| 47 |
+
resnet50v2.fc = nn.Sequential(
|
| 48 |
+
nn.Linear(num_ftrs, 500),
|
| 49 |
+
nn.ReLU(),
|
| 50 |
+
nn.Linear(500, 200),
|
| 51 |
+
nn.Dropout(0.5),
|
| 52 |
+
nn.Linear(200,40),
|
| 53 |
+
nn.ReLU(),
|
| 54 |
+
nn.Dropout(0.3),
|
| 55 |
+
nn.Linear(40,10),
|
| 56 |
+
nn.ReLU()
|
| 57 |
+
|
| 58 |
+
)
|
| 59 |
+
print("Model Info:")
|
| 60 |
+
print("ResNet50,Pretrained,weight adj. LR=0.01,Mom=0.3,WD=0.0001")
|
| 61 |
+
print("Schedule step=1,gamma=0.7, 20 epoches")
|
| 62 |
+
|
| 63 |
+
# Move the model to the GPU
|
| 64 |
+
resnet50v2 = resnet50v2.to(device, dtype=torch.float32)
|
| 65 |
+
|
| 66 |
+
|
| 67 |
+
optimizer = torch.optim.SGD(resnet50v2.parameters(), lr=0.01,momentum=0.3,weight_decay=0.0001)
|
| 68 |
+
criterion = nn.CrossEntropyLoss()
|
| 69 |
+
scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=5, gamma=0.5)
|
| 70 |
+
train_losses = []
|
| 71 |
+
test_losses = []
|
| 72 |
+
accuracies = []
|
| 73 |
+
train_acc = []
|
| 74 |
+
|
| 75 |
+
for epoch in range(20): # loop over the dataset multiple times
|
| 76 |
+
running_loss = 0.0
|
| 77 |
+
correctTrain = 0
|
| 78 |
+
totalTrain = 0
|
| 79 |
+
pbar = tqdm(enumerate(trainloaderRes16, 0), total=len(trainloaderRes16), desc="Epoch {}".format(epoch+1))
|
| 80 |
+
for i, data in pbar:
|
| 81 |
+
# get the inputs; data is a list of [inputs, labels]
|
| 82 |
+
inputs, labels = data[0].to(device,dtype=torch.float32), data[1].to(device)
|
| 83 |
+
|
| 84 |
+
# zero the parameter gradients
|
| 85 |
+
optimizer.zero_grad()
|
| 86 |
+
|
| 87 |
+
# forward + backward + optimize
|
| 88 |
+
outputs = resnet50v2(inputs)
|
| 89 |
+
loss = criterion(outputs, labels)
|
| 90 |
+
loss.backward()
|
| 91 |
+
optimizer.step()
|
| 92 |
+
|
| 93 |
+
# print statistics
|
| 94 |
+
running_loss += loss.item()
|
| 95 |
+
_, predicted_train = torch.max(outputs.data, 1)
|
| 96 |
+
totalTrain += labels.size(0)
|
| 97 |
+
correctTrain += (predicted_train == labels).sum().item()
|
| 98 |
+
pbar.set_postfix({'loss': running_loss/(i+1)})
|
| 99 |
+
train_accuracy = 100 * correctTrain / totalTrain
|
| 100 |
+
train_acc.append(train_accuracy)
|
| 101 |
+
print(f'Epoch {epoch + 1} loss: {running_loss / len(trainloaderRes16):.3f}')
|
| 102 |
+
|
| 103 |
+
# Start of testing phase
|
| 104 |
+
resnet50v2.eval() # Set the model to evaluation mode
|
| 105 |
+
test_loss = 0.0
|
| 106 |
+
correct = 0
|
| 107 |
+
total = 0
|
| 108 |
+
with torch.no_grad():
|
| 109 |
+
for data in testloaderRes16:
|
| 110 |
+
images, labels = data[0].to(device,dtype=torch.float32), data[1].to(device)
|
| 111 |
+
outputs = resnet50v2(images)
|
| 112 |
+
loss = criterion(outputs, labels)
|
| 113 |
+
test_loss += loss.item()
|
| 114 |
+
_, predicted = torch.max(outputs.data, 1)
|
| 115 |
+
total += labels.size(0)
|
| 116 |
+
correct += (predicted == labels).sum().item()
|
| 117 |
+
|
| 118 |
+
print(f'Epoch {epoch + 1} Test loss: {test_loss / len(testloaderRes16):.3f}, Accuracy: {100 * correct / total:.2f}%')
|
| 119 |
+
#print the learning rate
|
| 120 |
+
print(f'Epoch {epoch + 1} Learning rate: {optimizer.param_groups[0]["lr"]}')
|
| 121 |
+
train_losses.append(running_loss / len(trainloaderRes16))
|
| 122 |
+
test_losses.append(test_loss / len(testloaderRes16))
|
| 123 |
+
accuracies.append(100 * correct / total)
|
| 124 |
+
|
| 125 |
+
resnet50v2.train() # Set the model back to training model
|
| 126 |
+
scheduler.step()
|
| 127 |
+
|
| 128 |
+
|
| 129 |
+
|
| 130 |
+
print('Finished Training')
|
| 131 |
+
|
| 132 |
+
plt.figure(figsize=(10, 5))
|
| 133 |
+
plt.plot(train_losses, label='Training Loss')
|
| 134 |
+
plt.plot(test_losses, label='Test Loss')
|
| 135 |
+
plt.xlabel('Epochs')
|
| 136 |
+
plt.ylabel('Loss')
|
| 137 |
+
plt.legend()
|
| 138 |
+
plt.show()
|
| 139 |
+
|
| 140 |
+
plt.figure(figsize=(10, 5))
|
| 141 |
+
plt.plot(accuracies, label='Accuracy')
|
| 142 |
+
plt.plot(train_acc, label='Training Accuracy')
|
| 143 |
+
plt.xlabel('Epochs')
|
| 144 |
+
plt.ylabel('Accuracy (%)')
|
| 145 |
+
plt.legend()
|
| 146 |
+
plt.show()
|
| 147 |
+
```
|
| 148 |
+
|