Torch MNIST number detection example - sendtex
Training Model - Deep Learning and Neural Networks with Python and Pytorch
Example code for detecting numbers in MNIST dataset:
#!/usr/bin/python
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import torchvision
from torchvision import transforms, datasets
import matplotlib.pyplot as plt
train = datasets.MNIST("", train=True, download=True,
transform=transforms.Compose([transforms.ToTensor()]))
test = datasets.MNIST("", train=False, download=True,
transform=transforms.Compose([transforms.ToTensor()]))
trainset = torch.utils.data.DataLoader(train, batch_size=10, shuffle=True)
testset = torch.utils.data.DataLoader(test, batch_size=10, shuffle=True)
# How does data look like? Comment out the print
for data in trainset:
#print(data)
break
X, y = data[0][0], data[1][0]
#print(data[0][0].shape)
#plt.imshow(data[0][0].view(28,28))
#plt.show()
class Net(nn.Module):
def __init__(self):
super().__init__()
self.fc1 = nn.Linear(784, 64) # 784 = 28*28
self.fc2 = nn.Linear(64, 64)
self.fc3 = nn.Linear(64, 64)
self.fc4 = nn.Linear(64, 10)
def forward(self, x):
x = F.relu(self.fc1(x))
x = F.relu(self.fc2(x))
x = F.relu(self.fc3(x))
x = self.fc4(x)
return F.log_softmax(x, dim=1)
net = Net()
X = torch.rand(28,28)
X = X.view(-1, 28*28)
output = net(X)
#print(output) # displays output tensor
#print(torch.argmax(output)) # displays detected number
optimizer = optim.Adam(net.parameters(), lr=0.01)
EPOCHS = 3
for epoch in range(EPOCHS):
for data in trainset:
# data is batch of featuresets and labels
X, y = data
net.zero_grad()
output = net(X.view(-1, 28*28))
loss = F.nll_loss(output, y)
loss.backward()
optimizer.step()
print(loss)
correct = 0
total = 0
goodset = []
badset = []
with torch.no_grad():
for data in testset:
X, y = data
output = net(X.view(-1, 784))
for idx, i in enumerate(output):
if torch.argmax(i) == y[idx]:
correct += 1
goodset.append([X[idx], y[idx]])
else:
badset.append([X[idx], y[idx], torch.argmax(i)])
total += 1
print("Accuracy: ", round(correct/total, 3))
print("Not detected: ", len(failedset))
def display_good_image(i):
plt.imshow(goodset[i][0].view(28,28))
plt.show()
print("Image {} detected {}".format(goodset[i][1], goodset[i][1]))
def display_bad_image(i):
plt.imshow(badset[i][0].view(28,28))
plt.show()
print("Image {} detected {}".format(badset[i][1], badset[i][2]))
# Display 1st badly detected image. Close window to proceed.
display_bad_image(0)
# Display 1st correctly detected image. Close window to proceed.
display_good_image(0)