Skip to content

Machine-learning features

HabitabilityFeatureTransformer is a scikit-learn transformer. It turns raw NASA Exoplanet Archive columns into 15 habitability features. It needs pip install "exohab[ml]" or "exohab[all]".

import exohab as eh

df = eh.load_nasa_exoplanets()
t = eh.HabitabilityFeatureTransformer(surface_temp_method="kashyap2017")
X = t.fit_transform(df)
print(X.shape)        # (number of planets, 15)

Features

Feature Meaning
density_eu, escape_velocity_eu Derived from mass and radius
esi_2p, esi_4p_interior, esi_4p_surface, esi_4p ESI scores
esi_interior_surface_ratio Interior ESI divided by surface ESI
hz_seff_inner, hz_seff_outer Conservative habitable-zone flux limits
hz_seff_inner_optimistic, hz_seff_outer_optimistic Optimistic limits
hz_inner_au, hz_outer_au Conservative limits as distances
in_conservative_hz, in_optimistic_hz 1.0 inside, 0.0 outside, NaN if unknown

Habitable-zone features are NaN for stars outside 2600–7200 K, where the Kopparapu fits are not valid.

In a pipeline

import exohab as eh
from sklearn.cluster import KMeans
from sklearn.impute import SimpleImputer
from sklearn.pipeline import make_pipeline

df = eh.load_nasa_exoplanets()
model = make_pipeline(
    eh.HabitabilityFeatureTransformer(),
    SimpleImputer(strategy="median"),
    KMeans(n_clusters=4, n_init="auto", random_state=0),
)
df["cluster"] = model.fit_predict(df)

Options

  • Column names default to the NASA archive names (pl_rade, pl_bmasse, pl_insol, pl_eqt, st_teff, st_lum) and can all be changed.
  • keep_original=True returns your input columns followed by the features.
  • The input must be a pandas DataFrame.
  • The transformer is stateless: fit only checks the input.