Add evaluation configuration and update inference pipeline

Refactor configuration parameters and update model architecture references

Update learning rate scheduler parameters and clean up loss calculation notes

Rename localpatch.yaml to train_config.yaml and update notes; modify learning rate scheduler parameters and rename attention mask creation method

Updated model to include inverted attention naming scheme

Added checkpointing

major updates for flare catalog creation

Merge remote-tracking branch 'origin/main' into dev-patch

more changes

cleaning patch model

Various bug fixes and changes to all codes

enhance data splitting functionality.

made evaluation easier by adding auto eval... makes configs on its own...

Created a cnn model and local attention based model to get more physically significant predictions...

Enhance FlareDownloadProcessor to support time span mode for 1-minute cadence downloads, adding new arguments and methods for data existence checks. Update command-line arguments for improved usability. Refactor evaluation configurations and model training settings, including adjustments to checkpoint paths and model parameters for better performance.

Refactor ViT model configuration and base weights; update learning rate in training configs; standardize Huber loss delta notation for consistency.

Refactor SXRRegressionDynamicLoss weights and performance multipliers; adjust model configuration for ViTPatch with reduced patch size and increased number of heads; implement GPU memory isolation and monitoring in training script; enhance AttentionMapCallback to accept dynamic patch size.

add 2d positional embedding to try and improve forcasting

vit changes

Update data processing scripts to use new combined directory structure; adjust validation and training splits, enhance loss calculation with updated parameters, and modify configuration files for improved model training and evaluation.

Refactor SolarFlareEvaluator for baseline-only mode; update evaluation logic and configuration files. Adjusted model prediction handling, added baseline-only mode checks, and modified output paths in YAML configurations.

edited code slightly so that base weights are calculated based on number of samples in training (patch based)

Updated window size in SXRRegressionDynamicLoss to 1500, added forward_for_callback method in VisionTransformer for compatibility with AttentionMapCallback, and modified selected_model names in config files to 'ViTPatch'.

back to huber

updates to base model

Updated splitting function so now it can do SXR and AIA individually... it can also repartition an already partioned folder... updated the megs models to include warm restarts, weight decay, and adaptive weights

Experimenting With Fusion Model

bug fix

bug fixes

Fixed patch weights and loss to mse

updated tree structure