Barebones training pipeline

[cl676767]

Accept plz

[StealthHydra179]

Could you write a docstring at the start with the explanation on how to use the training pipeline?

[StealthHydra179]

Could you also write a simple test to make sure the code is working in the test folder?

[Copilot]

Pull Request Overview

This PR introduces a basic training pipeline for machine learning models using PyTorch. The implementation provides essential functionality for training, validation, and model saving.

• Implements a TrainingPipeline class with configurable hyperparameters
• Provides methods for training epochs, validation, and model persistence
• Sets up automatic train/validation data splitting and device management

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[Copilot]

The train/test split logic is incorrect. Setting train_size to the full length of data results in val_size being 0, meaning no validation data. For an 80/20 split, train_size should be int(0.8 * len(data)).

Suggested change

	train_size = int(len(data))
	train_size = int(0.8 * len(data))

[Copilot]

The arguments to CrossEntropyLoss are in the wrong order. It should be self.criterion(preds, yb) where predictions come first and targets second.

[Copilot]

Missing parentheses for the context manager. Should be with torch.inference_mode():.

Suggested change

	with torch.inference_mode:
	with torch.inference_mode():

[Copilot]

Incorrect attribute reference. Should be self.val_loader instead of self.val_data to iterate through the validation DataLoader.

Suggested change

	for xb,yb in self.val_data:
	for xb,yb in self.val_loader:

[Copilot]

The arguments to CrossEntropyLoss are in the wrong order. It should be self.criterion(preds, yb) where predictions come first and targets second.

[Copilot]

Using predict method is unconventional for PyTorch models. Standard practice is to call the model directly with self.model(xb) or use self.model.forward(xb).

[Copilot]

Using predict method is unconventional for PyTorch models. Standard practice is to call the model directly with self.model(xb) or use self.model.forward(xb).

[StealthHydra179]

Some quick small changes

[StealthHydra179]

Could we move this to a separate test file? (see how the camera module did it)

[StealthHydra179]

suggestion: move the 0.8 to be a default parameter in init in case we want to test different splits