Usage

The project should be executed in the following order:

1. Download and prepare the dataset.

2. Create the features and labels.

3. Train and evaluate the model.

4. Generate predictions.

Download and prepare the dataset

The dataset module downloads the dataset from Hugging Face, creates a local sample, removes incomplete rows, and saves the cleaned dataset.

Run:

python -m asteroid_impact.dataset

By default, the project downloads a sample of up to 500 rows.

To download the dataset again, use:

python -m asteroid_impact.dataset --force-download

The generated files are:

• data/raw/dataset.csv

• data/processed/dataset.csv

Create features

The features module creates the input features and the target labels.

Run:

python -m asteroid_impact.features

This command creates:

• data/processed/features.csv

• data/processed/labels.csv

Train the model

The train module trains and evaluates the neural network.

Run:

python -m asteroid_impact.modeling.train

This command creates:

• models/impact_probability_model.keras

• models/feature_scaler.joblib

• reports/model_metrics.csv

• reports/training_history.csv

Generate predictions

The predict module loads the trained model and generates predictions.

Run:

python -m asteroid_impact.modeling.predict

The predictions are saved as a CSV file in the processed data directory.