torch101_full.py

torch.manual_seed(42)

x_tensor = torch.from_numpy(x).float()
y_tensor = torch.from_numpy(y).float()

# Builds dataset with ALL data
dataset = TensorDataset(x_tensor, y_tensor)
# Splits randomly into train and validation datasets
train_dataset, val_dataset = random_split(dataset, [80, 20])
# Builds a loader for each dataset to perform mini-batch gradient descent
train_loader = DataLoader(dataset=train_dataset, batch_size=16)
val_loader = DataLoader(dataset=val_dataset, batch_size=20)

# Builds a simple sequential model
model = nn.Sequential(nn.Linear(1, 1)).to(device)
print(model.state_dict())

# Sets hyper-parameters
lr = 1e-1
n_epochs = 150

# Defines loss function and optimizer
loss_fn = nn.MSELoss(reduction='mean')
optimizer = optim.SGD(model.parameters(), lr=lr)

losses = []
val_losses = []
# Creates function to perform train step from model, loss and optimizer
train_step = make_train_step(model, loss_fn, optimizer)

# Training loop
for epoch in range(n_epochs):
    # Uses loader to fetch one mini-batch for training
    for x_batch, y_batch in train_loader:
        # NOW, sends the mini-batch data to the device
        # so it matches location of the MODEL
        x_batch = x_batch.to(device)
        y_batch = y_batch.to(device)
        # One stpe of training
        loss = train_step(x_batch, y_batch)
        losses.append(loss)
        
    # After finishing training steps for all mini-batches,
    # it is time for evaluation!
        
    # We tell PyTorch to NOT use autograd...
    # Do you remember why?
    with torch.no_grad():
        # Uses loader to fetch one mini-batch for validation
        for x_val, y_val in val_loader:
            # Again, sends data to same device as model
            x_val = x_val.to(device)
            y_val = y_val.to(device)
            
            # What is that?!
            model.eval()
            # Makes predictions
            yhat = model(x_val)
            # Computes validation loss
            val_loss = loss_fn(y_val, yhat)
            val_losses.append(val_loss.item())

print(model.state_dict())
print(np.mean(losses))
print(np.mean(val_losses))

This was really helpful.

This is really helpful!

Thank you for this really complete and helpfull tutorial !

This was actually very helpful, thanks!

Thank you !

How would you implement a standard scaler into this workflow? You want to scale after the train-test split because it should be fit only on training data, but would you do that for each individual batch? Pytorch must have a standard scaler for Dataset objects no?
Thanks for the tutorial!

this is the best PyTorch 101 I have encountered so far. Thank you very much!

Thank you so much for your great tutorial!!🤗

I used to live in TensorFlow world, and moved to fastai one 2 years ago.
These days, I've been wondering what's going on inside fastai...
After reading your blog post, I got a super solid foundation of pytorch!

Thanks a lot for the amazing blog.

Thanks for the great tutorial about the basics of Pytorch

This is such a a good, concise tutorial! it helped a lot. Thank you!

great tutorial, thanks. I have a question, what a boat inference codes for new data, we should first save our model and load and run on new data like validation codes, right? in the case of 2-fold cross validation with k=5 when save the models(5models for k=5), when I load and run, all the 5 model have the same accuracy. I dont know what is wrong ?