soleneulmer/bem

Random forest for exoplanets

BEM : beyond the exoplanet mass-radius relation with random forest

Predicting the radius of exoplanets based on its planetary and stellar parameters

Branca Edmée Marques

A portuguese scientist who worked on nuclear physics in France with Marie Curie

To install bem

Requirement: Python 3.7, Scikit-learn 0.20.3

pip install bem

or

git clone https://github.com/soleneulmer/bem.git
cd bem
python setup.py install

A simple decision tree

to predict exoplanet radius

How to run bem:

1. Load dataset and model

# Load exoplanet and solar system planets dataset
dataset = bem.load_dataset()
# Plot the dataset radius as a function of mass and equilibrium temperature
bem.plot_dataset(dataset)

# Build the random forest model and predict radius of the dataset
regr, y_test_predict, _, train_test_sets = bem.random_forest_regression(dataset)

2. Predict the radius of your planet

# Predict a new radius				    with error
# my_planet = [planetary_mass (Me, Mj)              or  my_planet = [planetary_mass, planetary mass error,
#              semi major axis (AU),                                semi major axis, semi major axis error,
#              eccentricity,					    eccentricity, eccentricity error,
#              stellar radius (Rsun),				    stellar radius, stellar radius error,
#              stellar effective temperature (K), 		    stellar effective temperature, teff error,
#              stellar mass (Msun)]				    stellar mass (Msun), stellar mass error]

radius, my_pred_planet = bem.predict_radius(my_planet=np.array([[1.63,
								 0.034,
                                                 		 0.02,
                                                 		 0.337,
                                                 		 3505.0,
                                                 		 0.342]]),
                        		    my_name=np.array(['GJ 357 b']),
                            		    regr=regr,
                            		    jupiter_mass=False,
					    error_bar=False)
# If error_bar is True
# print('Radius: ', radius[0][0], '+-', radius[1])

3. Compute error bars for the radius predictions

# Load exoplanet and solar system planets dataset with uncertainties
dataset_errors = bem.load_dataset_errors()
# Compute the error bars for the test set planets
radii_test_output_error, _ = bem.computing_errorbars(regr,
                                                     dataset_errors,
                                                     train_test_sets)
# Plot the test set, true radius versus RF predicted radius
bem.plot_true_predicted(train_test_sets,
                        y_test_predict,
                        radii_test_output_error)

4. Radial velocity dataset

# Load the radial velocity dataset
dataset_rv = bem.load_dataset_RV()
# Predict the radius of the RV dataset
radii_RV_RF = regr.predict(dataset_rv)
# Plot the predictions of the RV dataset
bem.plot_dataset(dataset_rv, predicted_radii=radii_RV_RF, rv=True)

5. Diagnostic plots

# Plot the learning curve
bem.plot_learning_curve(regr, dataset)
# Plot the validation curves
bem.plot_validation_curves(regr, dataset, name='features')
bem.plot_validation_curves(regr, dataset, name='tree')
bem.plot_validation_curves(regr, dataset, name='depth')

6. LIME explanations

see their github

# Explain the RF predictions
# of the exoplanets from the test set
bem.plot_LIME_predictions(regr, dataset, train_test_sets)
# LIME explanation for your planet
# in this case GJ 357 b
bem.plot_LIME_predictions(regr, dataset, train_test_sets,
                          my_pred_planet=my_pred_planet,
                          my_true_radius=1.166)