retrain the classifier model

[drjova]

retrain classifier following the guide https://inspire.docs.cern.ch/operations/classifier/ and using the new PC

Implementation

• [x] Implement previous split for Training, Test and Validation Set
  • [x] Use create_dataset.py to create Training and Test split
  • [x] Use Dataloaders in classifier to split for training and validation set
  • [x] Adjust path structure to new splits
• [x] Implement ULMfit architecture
  • [x] Language Model implementation
    • [x] Create Vocabulary using min and max vocab size
    • [x] Load pre-trained Language Model
    • [x] Fine-tune pre-trained LM with Inspire Data
    • [x] Save fine-tuned LM encoder layer
  • [x] Classifier implementation
    • [x] Load classifier model
    • [x] Replace encoder layer
    • [x] Load and split Inspire data into validation and training set
    • [x] Train using Inspire Data
    • [x] Implement predict method
• [x] Cleanup unnecessary code from old implementation
• [x] Guarantee that the cli is useable as before (get all tests running)
• [ ] Fine-tune parameters for best results (also try one to one copy of old parameters)
• [x] Write readme with step by step instructions on how to use the classifier

Training

• [x] Pull training data for last x Years
• [x] Train classifier model with the new data

[drjova]

@PascalEgn could you please add the checklist?