Fix torch.load device handling and remove deprecated scheduler params

[Aryanjstar]

This pull request streamlines model loading and scheduler initialization across multiple anomaly detection and classification scripts. The main improvements include standardizing the use of torch.load with map_location=device and weights_only=True for loading model weights, and removing unnecessary verbosity from learning rate scheduler initialization. This results in cleaner, more maintainable, and device-agnostic code.

Model Loading Improvements:

• Replaced conditional logic for device-specific model loading with a unified approach using torch.load(MODEL_PATH, map_location=device, weights_only=True) for all models in both anomaly detection and classification pipelines. This change affects autoencoders, variational autoencoders, adversarial autoencoders, and classification models, ensuring consistent and robust loading behavior regardless of device. [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14]

Scheduler Initialization Cleanup:

• Removed the verbose parameter from the initialization of CosineAnnealingWarmRestarts and OneCycleLR learning rate schedulers, simplifying their usage and reducing unnecessary output in both anomaly detection and classification training scripts. [1] [2] [3] [4]

These changes collectively make the codebase more maintainable and less error-prone, especially when running on different devices or in production environments.