Bounded Regression.py

import numpy as np
from sklearn.svm import SVR
from sklearn.pipeline import Pipeline
from sklearn.compose import TransformedTargetRegressor

from scipy.stats import beta
from scipy.special import expit, logit

np.random.seed(0)
# Generate features
X = np.random.normal(size=(1000, 5))
# Combine them and ensure they are between 0 and 1
betas = np.random.normal(0, 1, size=5)
eta = 1 / (1 + np.exp(-X @ betas))
# Make outcomes
y = beta(eta * 50, (1 - eta) * 50).rvs()

plt.scatter(eta, y)

no_trans_pipe = Pipeline([("reg", SVR(gamma="scale"))])


trans_pipe = Pipeline(
    [
        (
            "reg",
            TransformedTargetRegressor(
                SVR(gamma="scale"), func=logit, inverse_func=expit
            ),
        )
    ]
)

Xtrain, Xtest, ytrain, ytest = train_test_split(X, y, train_size=0.5)


# Uh oh, predictions out of bounds!
(no_trans_pipe.fit(Xtrain, ytrain).predict(Xtest) < 0).any()

# Predictions stay constrained!
(trans_pipe.fit(Xtrain, ytrain).predict(Xtest) < 0).any() | (
    trans_pipe.fit(Xtrain, ytrain).predict(Xtest) > 1
).any()