MNIST dataset training (III)
Accuracy: training vs test data sets
In [1]:
import torch
from torchvision import datasets, transforms
# Define a transform to normalize the data
transform = transforms.Compose([transforms.ToTensor(),
#transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
transforms.Normalize((0.5,), (0.5,)),
])
# Download and load the training data
trainset = datasets.FashionMNIST('~/.pytorch/F_MNIST_data/', download=True, train=True, transform=transform)
trainloader = torch.utils.data.DataLoader(trainset, batch_size=64, shuffle=True)
# Download and load the test data
testset = datasets.FashionMNIST('~/.pytorch/F_MNIST_data/', download=True, train=False, transform=transform)
testloader = torch.utils.data.DataLoader(testset, batch_size=64, shuffle=True)
In [2]:
from torch import nn, optim
import torch.nn.functional as F
class Classifier(nn.Module):
def __init__(self):
super().__init__()
self.fc1 = nn.Linear(784, 256)
self.fc2 = nn.Linear(256, 128)
self.fc3 = nn.Linear(128, 64)
self.fc4 = nn.Linear(64, 10)
def forward(self, x):
# make sure input tensor is flattened
x = x.view(x.shape[0], -1)
x = F.relu(self.fc1(x))
x = F.relu(self.fc2(x))
x = F.relu(self.fc3(x))
x = F.log_softmax(self.fc4(x), dim=1)
return x
In [4]:
model = Classifier()
criterion = nn.NLLLoss()
optimizer = optim.Adam(model.parameters(), lr=0.003)
# Use GPU if it's available
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print("using:",device)
#device="cpu"
epochs = 10 #30
steps = 0
train_losses, test_losses = [], []
for e in range(epochs):
print(e)
running_loss = 0
model.to(device)
for images, labels in trainloader:
images, labels = images.to(device), labels.to(device)
optimizer.zero_grad()
log_ps = model(images)
loss = criterion(log_ps, labels)
loss.backward()
optimizer.step()
running_loss += loss.item()
else:
test_loss = 0
accuracy = 0
# Turn off gradients for validation, saves memory and computations
with torch.no_grad():
model.to('cpu')
images, labels = images.to('cpu'), labels.to('cpu')
for images, labels in testloader:
log_ps = model(images)
test_loss += criterion(log_ps, labels)
ps = torch.exp(log_ps)
top_p, top_class = ps.topk(1, dim=1)
equals = top_class == labels.view(*top_class.shape)
accuracy += torch.mean(equals.type(torch.FloatTensor))
train_losses.append(running_loss/len(trainloader))
test_losses.append(test_loss/len(testloader))
print("Epoch: {}/{}.. ".format(e+1, epochs),
"Training Loss: {:.3f}.. ".format(running_loss/len(trainloader)),
"Test Loss: {:.3f}.. ".format(test_loss/len(testloader)),
"Test Accuracy: {:.3f}".format(accuracy/len(testloader)))
using: cuda 0 Epoch: 1/10.. Training Loss: 0.519.. Test Loss: 0.436.. Test Accuracy: 0.840 1 Epoch: 2/10.. Training Loss: 0.396.. Test Loss: 0.418.. Test Accuracy: 0.842 2 Epoch: 3/10.. Training Loss: 0.356.. Test Loss: 0.394.. Test Accuracy: 0.864 3 Epoch: 4/10.. Training Loss: 0.337.. Test Loss: 0.375.. Test Accuracy: 0.865 4 Epoch: 5/10.. Training Loss: 0.317.. Test Loss: 0.366.. Test Accuracy: 0.871 5 Epoch: 6/10.. Training Loss: 0.304.. Test Loss: 0.392.. Test Accuracy: 0.862 6 Epoch: 7/10.. Training Loss: 0.289.. Test Loss: 0.351.. Test Accuracy: 0.878 7 Epoch: 8/10.. Training Loss: 0.284.. Test Loss: 0.390.. Test Accuracy: 0.869 8 Epoch: 9/10.. Training Loss: 0.272.. Test Loss: 0.371.. Test Accuracy: 0.874 9 Epoch: 10/10.. Training Loss: 0.265.. Test Loss: 0.379.. Test Accuracy: 0.870
In [5]:
%matplotlib inline
%config InlineBackend.figure_format = 'retina'
import matplotlib.pyplot as plt
plt.plot(train_losses, label='Training loss')
plt.plot(test_losses, label='Validation loss')
plt.legend(frameon=False)
Out[5]:
<matplotlib.legend.Legend at 0x7fe6c43d9a90>
see part 5
In [10]:
class Classifier(nn.Module):
def __init__(self):
super().__init__()
self.fc1 = nn.Linear(784, 256)
self.fc2 = nn.Linear(256, 128)
self.fc3 = nn.Linear(128, 64)
self.fc4 = nn.Linear(64, 10)
# Dropout module with 0.2 drop probability
self.dropout = nn.Dropout(p=0.2)
def forward(self, x):
# make sure input tensor is flattened
x = x.view(x.shape[0], -1)
# Now with dropout
x = self.dropout(F.relu(self.fc1(x)))
x = self.dropout(F.relu(self.fc2(x)))
x = self.dropout(F.relu(self.fc3(x)))
# output so no dropout here
x = F.log_softmax(self.fc4(x), dim=1)
return x
In [11]:
model = Classifier()
criterion = nn.NLLLoss()
optimizer = optim.Adam(model.parameters(), lr=0.003)
epochs = 30
steps = 0
train_losses, test_losses = [], []
for e in range(epochs):
model.to(device)
running_loss = 0
for images, labels in trainloader:
images, labels = images.to(device), labels.to(device)
optimizer.zero_grad()
log_ps = model(images)
loss = criterion(log_ps, labels)
loss.backward()
optimizer.step()
running_loss += loss.item()
else:
test_loss = 0
accuracy = 0
# Turn off gradients for validation, saves memory and computations
with torch.no_grad():
model.to('cpu')
images, labels = images.to('cpu'), labels.to('cpu')
model.eval()
for images, labels in testloader:
log_ps = model(images)
test_loss += criterion(log_ps, labels)
ps = torch.exp(log_ps)
top_p, top_class = ps.topk(1, dim=1)
equals = top_class == labels.view(*top_class.shape)
accuracy += torch.mean(equals.type(torch.FloatTensor))
model.train()
train_losses.append(running_loss/len(trainloader))
test_losses.append(test_loss/len(testloader))
print("Epoch: {}/{}.. ".format(e+1, epochs),
"Training Loss: {:.3f}.. ".format(train_losses[-1]),
"Test Loss: {:.3f}.. ".format(test_losses[-1]),
"Test Accuracy: {:.3f}".format(accuracy/len(testloader)))
Epoch: 1/30.. Training Loss: 0.605.. Test Loss: 0.468.. Test Accuracy: 0.830 Epoch: 2/30.. Training Loss: 0.477.. Test Loss: 0.478.. Test Accuracy: 0.827 Epoch: 3/30.. Training Loss: 0.451.. Test Loss: 0.407.. Test Accuracy: 0.853 Epoch: 4/30.. Training Loss: 0.432.. Test Loss: 0.410.. Test Accuracy: 0.855 Epoch: 5/30.. Training Loss: 0.420.. Test Loss: 0.393.. Test Accuracy: 0.860 Epoch: 6/30.. Training Loss: 0.409.. Test Loss: 0.411.. Test Accuracy: 0.856 Epoch: 7/30.. Training Loss: 0.409.. Test Loss: 0.396.. Test Accuracy: 0.861 Epoch: 8/30.. Training Loss: 0.400.. Test Loss: 0.400.. Test Accuracy: 0.864 Epoch: 9/30.. Training Loss: 0.398.. Test Loss: 0.433.. Test Accuracy: 0.850 Epoch: 10/30.. Training Loss: 0.390.. Test Loss: 0.380.. Test Accuracy: 0.865 Epoch: 11/30.. Training Loss: 0.381.. Test Loss: 0.385.. Test Accuracy: 0.868 Epoch: 12/30.. Training Loss: 0.377.. Test Loss: 0.400.. Test Accuracy: 0.861 Epoch: 13/30.. Training Loss: 0.377.. Test Loss: 0.403.. Test Accuracy: 0.856 Epoch: 14/30.. Training Loss: 0.372.. Test Loss: 0.391.. Test Accuracy: 0.866 Epoch: 15/30.. Training Loss: 0.370.. Test Loss: 0.411.. Test Accuracy: 0.864 Epoch: 16/30.. Training Loss: 0.376.. Test Loss: 0.392.. Test Accuracy: 0.866 Epoch: 17/30.. Training Loss: 0.365.. Test Loss: 0.361.. Test Accuracy: 0.870 Epoch: 18/30.. Training Loss: 0.370.. Test Loss: 0.367.. Test Accuracy: 0.871 Epoch: 19/30.. Training Loss: 0.361.. Test Loss: 0.376.. Test Accuracy: 0.864 Epoch: 20/30.. Training Loss: 0.359.. Test Loss: 0.379.. Test Accuracy: 0.868 Epoch: 21/30.. Training Loss: 0.360.. Test Loss: 0.380.. Test Accuracy: 0.864 Epoch: 22/30.. Training Loss: 0.351.. Test Loss: 0.365.. Test Accuracy: 0.871 Epoch: 23/30.. Training Loss: 0.349.. Test Loss: 0.385.. Test Accuracy: 0.869 Epoch: 24/30.. Training Loss: 0.361.. Test Loss: 0.380.. Test Accuracy: 0.868 Epoch: 25/30.. Training Loss: 0.347.. Test Loss: 0.376.. Test Accuracy: 0.871 Epoch: 26/30.. Training Loss: 0.342.. Test Loss: 0.378.. Test Accuracy: 0.867 Epoch: 27/30.. Training Loss: 0.344.. Test Loss: 0.384.. Test Accuracy: 0.868 Epoch: 28/30.. Training Loss: 0.347.. Test Loss: 0.372.. Test Accuracy: 0.872 Epoch: 29/30.. Training Loss: 0.340.. Test Loss: 0.382.. Test Accuracy: 0.872 Epoch: 30/30.. Training Loss: 0.350.. Test Loss: 0.388.. Test Accuracy: 0.868
In [12]:
%matplotlib inline
%config InlineBackend.figure_format = 'retina'
import matplotlib.pyplot as plt
plt.plot(train_losses, label='Training loss')
plt.plot(test_losses, label='Validation loss')
plt.legend(frameon=False)
Out[12]:
<matplotlib.legend.Legend at 0x7fe6eb045c18>
In [13]:
#model.state_dict().keys()
torch.save(model.state_dict(), 'checkpoint1-dropout.pth')
In [ ]:
#state_dict = torch.load('checkpoint1-dropout.pth')
#print(state_dict.keys())
#model.load_state_dict(state_dict)
Test
In [14]:
# Import helper module (should be in the repo)
import helper
# Test out your network!
model.eval()
dataiter = iter(testloader)
images, labels = dataiter.next()
img = images[0]
# Convert 2D image to 1D vector
img = img.view(1, 784)
# Calculate the class probabilities (softmax) for img
with torch.no_grad():
output = model.forward(img)
ps = torch.exp(output)
# Plot the image and probabilities
helper.view_classify(img.view(1, 28, 28), ps, version='Fashion')
In [ ]: