Lasso example

lasso.py

import numpy as np
from sklearn.linear_model import Lasso, LinearRegression

# Generate 100 samples of y = x**2 plus some small random error
n = 100
x = np.random.random(100)
y = x**2 + 0.1*np.random.random(n)
# Create [x,x**2] as different features
f = np.asarray([x,x**2]).T
# Train a linear regression model as comparison
m = LinearRegression()
# Fit the model
m.fit(f,y)
# Get cooefficients
print(m.coef_) # [0.1, 0.1, 0.1, 0.1, 0.1]

# Try to add more and more l1-regularization
# to make the solution more sparse.
# A linear model can then be trained on the remaining features
for alpha in 10**np.arange(-9,0):
    # Initialize the lasso class
    m = Lasso(alpha=alpha)
    m.fit(f,y)
    print(alpha, m.coef_)