qreg

qreg is a Python library for data sparse and non-parametric quantile regression. It implements quantile regression with matrix-valued kernels and makes it possible to learn several quantile curves simultaneously with a sparsity requirement on supporting data.

Highlights:

based on the library lightning;
follows the scikit-learn style of programming;
computationally demanding parts implemented in Cython.

Example

import numpy as np
import matplotlib.pyplot as plt
from qreg import QRegressor, toy_data

# Quantile levels to prediect
probs = np.linspace(0.1, 0.9, 5)

# Train and test dataset
x_train, y_train, z_train = toy_data(50)
x_test, y_test, z_test = toy_data(1000, t_min=-0.2, t_max=1.7, probs=probs)

# Define the quantile regressor
reg = QRegressor(C=1e2,  # Trade-off parameter
                 probs=probs,  # Quantile levels
                 gamma_out=1e-2,  # Inner kernel parameter
                 eps=2,  # Epsilon-loss level
                 alg='sdca',  # Algorithm (can change to 'qp')
                 max_iter=1e4,  # Maximal number of iteration
                 active_set=True,  # Active set strategy
                 verbose=True)

# Fit on training data and predict on test data
reg.fit(x_train, y_train)
pred = reg.predict(x_test)

# Plot the estimated conditional quantiles
plt.plot(x_train, y_train, '.')
for q in pred:
    plt.plot(x_test, q, '-')
for q in z_test:
    plt.plot(x_test, q, '--')

# Print some information
print("Objective value: %f" % reg.obj)
print("Training time: %0.2fs" % reg.time)
print("#SV: %d" % reg.num_sv())
print("Score: %f" % reg.score(x_test, y_test))

plt.show()

Dependencies

qreg needs Python >= 2.7, setuptools, Numpy, SciPy, scikit-learn, cvxopt and a working C/C++ compiler.

Installation

To install qreg from pip, type:

pip install https://github.com/msangnier/qreg/archive/master.zip

To install qreg from source, type:

git clone https://github.com/msangnier/qreg.git
cd qreg
python setup.py build
sudo python setup.py install

Authors

Olivier Fercoq and Maxime Sangnier

References

Data sparse nonparametric regression with epsilon-insensitive losses (2017), M. Sangnier, O. Fercoq, F. d'Alché-Buc. Asian Conference on Machine Learning (ACML).
Joint quantile regression in vector-valued RKHSs (2016), M. Sangnier, O. Fercoq, F. d'Alché-Buc. Neural Information Processing Systems (NIPS).

np-8 / qreg