Spaces:
Runtime error
Runtime error
File size: 6,069 Bytes
569aa5e e4b9ec4 a232c52 e4b9ec4 569aa5e e4b9ec4 569aa5e |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 |
import os
import time
import torch
import torch.nn as nn
import torchvision.datasets as datasets
import torchvision.transforms as transforms
from torch.utils.data import DataLoader
import torch.optim as optim
from tqdm import tqdm
def get_mean_std(loader):
'''
Calculates mean and std of input images.
Args:
loader (torch.DataLoader): Loader with images
Returns:
mean (torch.Tensor): Mean of images in loader
std (torch.Tensor): Standard deviation of images in loader
'''
channels_sum, channels_squared_sum, num_batches = 0, 0, 0
for data, _ in loader:
channels_sum += torch.mean(data, dim=[0,2,3]) # mean across [no. of examples, height, width]
channels_squared_sum += torch.mean(data**2, dim=[0,2,3]) # squared mean across [no. of examples, height, width]
num_batches += 1
mean = channels_sum/num_batches
std = (channels_squared_sum/(num_batches-mean**2))**0.5
return mean, std
class Net(nn.Module):
'''
model definition
'''
def __init__(self):
super(Net, self).__init__()
self.layer1 = nn.Sequential(
nn.Conv2d(1, 32, kernel_size=5),
nn.ReLU(),
)
self.layer2 = nn.Sequential(
nn.Conv2d(32, 32, kernel_size=5, bias=False),
nn.BatchNorm2d(32),
nn.ReLU(),
nn.MaxPool2d((2, 2)),
nn.Dropout2d(0.25),
)
self.layer3 = nn.Sequential(
nn.Conv2d(32, 64, kernel_size=3),
nn.ReLU(),
)
self.layer4 = nn.Sequential(
nn.Conv2d(64, 64, kernel_size=3, bias=False),
nn.BatchNorm2d(64),
nn.ReLU(),
nn.MaxPool2d((2, 2)),
nn.Dropout2d(0.25),
nn.Flatten(),
)
self.layer5 = nn.Sequential(
nn.Linear(576, 256, bias=False),
nn.BatchNorm1d(256),
nn.ReLU(),
)
self.layer6 = nn.Sequential(
nn.Linear(256, 128, bias=False),
nn.BatchNorm1d(128),
nn.ReLU(),
)
self.layer7 = nn.Sequential(
nn.Linear(128, 84, bias=False),
nn.BatchNorm1d(84),
nn.ReLU(),
nn.Dropout(0.25),
)
self.layer8 = nn.Sequential(
nn.Linear(84, 10),
nn.LogSoftmax(dim=1),
)
def forward(self, x):
x = transforms.Normalize(mean, std)(x)
x = self.layer1(x)
x = self.layer2(x)
x = self.layer3(x)
x = self.layer4(x)
x = self.layer5(x)
x = self.layer6(x)
x = self.layer7(x)
x = self.layer8(x)
return x
# downloads and loads MNIST train set
transform = transforms.Compose([transforms.ToTensor(), transforms.RandomAffine(degrees=10, translate=(0.1,0.1))])
train_data = datasets.MNIST(root='./data', train=True, download=True, transform=transform)
train_loader = DataLoader(dataset=train_data, batch_size=64, shuffle=True, pin_memory=True)
# downloads and loads MNIST test set
val_data = datasets.MNIST(root='./data', train=False, download=True, transform=transforms.ToTensor())
val_loader = DataLoader(dataset=train_data, batch_size=64, shuffle=False, pin_memory=True)
# uses GPU if available
if torch.cuda.is_available():
dev = "cuda:0"
else:
dev = "cpu"
device = torch.device(dev)
# gets mean and std of dataset
mean, std = get_mean_std(train_loader)
def run_model():
# defines parameters
model = Net().to(device=device)
optimizer = optim.Adam(model.parameters(), lr=0.1)
scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer, factor=0.2, patience=2)
criterion = nn.NLLLoss()
# iterates through epochs
for epoch in range(30):
print(f"\nEpoch {epoch+1}/{30}.")
# train loop
model.train()
total_train_loss = 0
total_correct = 0
for i, (images, labels) in enumerate(tqdm(train_loader)):
images = images.to(device)
labels = labels.to(device)
optimizer.zero_grad()
outputs = model(images)
loss = criterion(outputs, labels)
total_train_loss += loss.item()
loss.backward()
optimizer.step()
# Calculates train accuracy
outputs_probs = nn.functional.softmax(
outputs, dim=1) # gets probabilities
for idx, preds in enumerate(outputs_probs):
# if label with max probability matches true label
if labels[idx] == torch.argmax(preds.data):
total_correct += 1
train_loss = total_train_loss/(i+1)
train_accuracy = total_correct/len(train_data)
print(f"Train set:- Loss: {train_loss}, Accuracy: {train_accuracy}.")
# saves model state
if not os.path.exists("./saved_models"):
os.mkdir("./saved_models")
torch.save(model.state_dict(), f"./saved_models/mnist-cnn-{time.time()}.pt")
# val loop
model.eval()
total_val_loss = 0
total_correct = 0
with torch.no_grad():
for i, (images, labels) in enumerate(tqdm(val_loader)):
images = images.to(device)
labels = labels.to(device)
outputs = model(images)
loss = criterion(outputs, labels)
total_val_loss += loss.item()
outputs_probs = nn.functional.softmax(outputs, dim=1)
for idx, preds in enumerate(outputs_probs):
if labels[idx] == torch.argmax(preds.data):
total_correct += 1
val_loss = total_val_loss/(i+1)
val_accuracy = total_correct/len(val_data)
print(f"Val set:- Loss: {val_loss}, Accuracy: {val_accuracy}.")
# adjusts lr
scheduler.step(val_loss)
|