Update code to work with datasets shared in #34

[gabrieltseng]

• [x] Use the val .txt file to create the train val split
• [x] Update catboost_ -> worldcereal_
• [x] Update pixelids -> sample_id

Run: https://wandb.ai/nasa-harvest/presto-worldcereal/runs/zwolm9c6. Results with all the data (from this run):

Model	F1 score	Precision	Recall
CatBoost	0.849	0.846	0.852
Presto finetuning	0.857	0.813	0.907
Presto finetuning + RF	0.866	0.894	0.839
Presto finetuning + LR	0.859	0.812	0.911
Presto finetuning + CatBoost	0.863	0.822	0.908

[kvantricht]

@gabrieltseng Note that I uploaded a new file to the OneDrive containing the list of sample_ids that contain the original VAL set. You could use those to replicate the previous split.