Model Card: AutoML Tabular Predictor for HW2
Purpose
This model was trained as part of Homework 2 to explore the use of AutoML (AutoGluon Tabular) on a tabular dataset created by a classmate. The goal was to automatically search for strong models and hyperparameters to solve a binary classification problem on a small, augmented dataset.
Dataset
- Source Dataset: emkessle/TabularDataHW1
- Target Column:
"Target (Composer)"
- Splits:
- Augmented: synthetic data (300 rows)
- Original: smaller validation dataset (30 rows)
- Preprocessing:
- Dropped rows with missing values.
- Train/test split (80%/20%) stratified on target column.
Training
- Framework: AutoGluon Tabular
- Preset:
"best_quality"
- Time Limit: 300 seconds
- Explored Models: LightGBM, CatBoost, Random Forest, Neural Net, KNN, and ensembles.
Results
- Validation Split:
- Accuracy: 1.000
- Weighted F1: 1.000
Intended Use & Limits
- Intended Use: Educational use for class assignment.
- Not for Production: Model is not validated for real-world deployment.
- Limitations:
- Dataset is small and partially synthetic.
- Predictions are not generalizable.
- Labels are based on survey-like responses and may not reflect broader populations.
Ethical Considerations
- Dataset is classroom-generated, not representative of a real population.
- Augmented data may contain synthetic biases.
- Predictions should not be used in decision-making contexts.
License
This model is released under the MIT License for educational purposes.
AI Usage Disclosure
- Training relied on AutoML (AutoGluon Tabular) for model selection and hyperparameter tuning.
- Hugging Face Hub is used for hosting.
- No manual tuning or custom architectures were introduced.
- ChatGPT used to draft this model card
