README

Project Structure

.
├── final_data/              # Contains preprocessed data
│   ├── test_df.pkl         # Preprocessed test data
│   └── train_df.pkl        # Preprocessed train data
├── final_features/         # Feature csvs
│   ├── 512_RoBERTa_attention/
│   ├── gemini_attention/
│   ├── period_longformer_att/
│   ├── time_series/
│   └── word_attention_glove/
├── models/                 # Models Stored here (empty now)
└── *.ipynb                # Jupyter notebooks for different components

Running the Code

Prerequisites

• Python 3.x
• Required packages (Python, TensorFlow, PyTorch, xgboost, catboost, Transformers)

Data Preparation

The preprocessed data is stored in final_data directory in pickle format (.pkl). If you need to preprocess raw data:

1. Open preprocessing.ipynb
2. Modify the following variables to point to your raw data:
   • train_path: Path to your training CSV file
   • test_path: Path to your test CSV file
3. Run the notebook to generate the preprocessed data files

Model Training and Evaluation

All metaclassifier notebooks are self-contained and can be run directly from their respective locations. They expect the preprocessed data to be in the final_features directory.

Available metaclassifiers:

• MetaClassifier_CatBoost.ipynb
• MetaClassifier_LSTM.ipynb
• MetaClassifier_TabNet_PCA.ipynb
• MetaClassifier_TabNet.ipynb
• MetaClassifier_XGBoost.ipynb

Feature Engineering

Additional feature engineering notebooks are provided:

• feature_engineering.ipynb
• 512_RoBERTA_attention.ipynb
• period_gemini_att.ipynb
• period_longformer_attention.ipynb
• word_glove_attention.ipynb
• gemini_embed_generator.ipynb

Each notebook is self contained and can be run independently as long as the preprocessed data is available in the expected location. For generating gemini embeddings an API key is required.