device = torch.device(“cpu”)# if torch.cuda.is_available() else torch.device(“cpu”)
device
epochs=10
train = pd.read_csv(‘train.csv’)
test = pd.read_csv(‘test.csv’)
def ImageData(raw: pd.DataFrame):
y = raw[‘label’].values
y.resize(y.shape[0],1)
x = raw[[i for i in raw.columns if i != ‘label’]].values
x = x.reshape([-1,1, 28, 28])
y = y.astype(int).reshape(-1)
x = x.astype(float)
return x, y
Convert to One Hot Encoding
def one_hot_embedding(labels, num_classes=10):
y = torch.eye(num_classes)
return y[labels]
x_train, y_train = ImageData(train)
Normalization
mean = x_train.mean()
std = x_train.std()
x_train = (x_train-mean)/std
#x_val = (x_val-mean)/std
Numpy to Torch Tensor
x_train = torch.from_numpy(np.float32(x_train)).to(device)
y_train = torch.from_numpy(y_train.astype(np.long)).to(device)
y_train = one_hot_embedding(y_train)
#x_val = torch.from_numpy(np.float32(x_val))
#y_val = torch.from_numpy(y_val.astype(np.long))
Convert into Torch Dataset
train_ds = TensorDataset(x_train, y_train)
#val_ds = TensorDataset(x_val,y_val)
train_dl = DataLoader(train_ds, batch_size=64)
plt.ion() # interactive mode
data_transforms = {
‘train_tensor’: transforms.Compose([
transforms.RandomResizedCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
]),
‘val’: transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
]),
}
“”"train = pd.read_csv(‘train.csv’)
train_drop_label = train.drop(columns = ‘label’)
creating tensor from targets_df
train_tensor = torch.tensor(train_drop_label.values)
test = pd.read_csv(‘test.csv’)
test_tensor = torch.tensor(test.values)
“”"
device = torch.device(“cuda:0” if torch.cuda.is_available() else “cpu”)
def image_show(n):
number_frame = train.iloc[n, 1:].as_matrix()
number = train.iloc[n,0]
number_frame = number_frame.reshape(28, 28)
plt.imshow(number_frame)
plt.title('Image of number - '+str(number))
plt.axis(‘off’)
plt.show()
x = image_show(2)
Initialize weight with xavier_uniform
def init_weights(m):
if type(m) == nn.Linear:
torch.nn.init.xavier_uniform(m.weight)
m.bias.data.fill_(0.01)
Flatten Later
class Flatten(nn.Module):
def forward(self, input):
return input.view(input.size(0), -1)
Train the network and print accuracy and loss overtime
Train the network and print accuracy and loss overtime
def fit(train_dl, model, loss, optim, epochs=10):
model = model.to(device)
print(‘Epoch\tAccuracy\tLoss’)
accuracy_overtime = []
loss_overtime = []
for epoch in range(epochs):
avg_loss = 0
correct = 0
total=0
for x, y in train_dl: # Iterate over Data Loder
# Forward pass
#print(x)
yhat = model(x)
print(yhat)
l = loss(y, yhat)
#Metrics
avg_loss+=l.item()
# Backward pass
optim.zero_grad()
l.backward()
optim.step()
# Metrics
_, original = torch.max(y, 1)
_, predicted = torch.max(yhat.data, 1)
total += y.size(0)
correct = correct + (original == predicted).sum().item()
accuracy_overtime.append(correct/total)
loss_overtime.append(avg_loss/len(train_dl))
print(epoch,accuracy_overtime[-1], loss_overtime[-1], sep='\t')
return accuracy_overtime, loss_overtime
model_ft = models.vgg16(pretrained = True)
n_classes = 10
for params in model_ft.features.parameters():
params.require_grad = False
add your own last layer which has 2 output features
n_inputs = model_ft.classifier[6].in_features
last_layer = nn.Linear(n_inputs, 2)
model_ft.classifier[6] = last_layer
criterion = nn.CrossEntropyLoss()
Observe that all parameters are being optimized
optimizer_ft = optim.SGD(model_ft.parameters(), lr=0.001, momentum=0.9)
Decay LR by a factor of 0.1 every 7 epochs
exp_lr_scheduler = lr_scheduler.StepLR(optimizer_ft, step_size=7, gamma=0.1)
accuracy, loss = [],[]
plot_accuracy_loss(*fit(train_dl,model_ft, criterion,optimizer_ft,
epochs=20))
error: RuntimeError: Given groups=1, weight of size 64 3 3 3, expected input[64, 1, 28, 28] to have 3 channels, but got 1 channels instead.
Yes I now that we need to change from 3 to 1 but I tried converting rgb to gray it just doesnt help. Pleae help
