In [1]:

    

# Least Absolute Shrinkage and Selection Operator (LASSO)
# LASSO is similar to Ridge Regression in that it penalizes
# regression by an amount.


    

In [2]:

    

from sklearn.datasets import make_regression


    

In [3]:

    

reg_data, reg_target = make_regression(n_samples=200, n_features=500,
                                      n_informative=5, noise=5)


    

In [10]:

    

from sklearn.linear_model import Lasso
import numpy as np


    

In [11]:

    

lasso = Lasso()


    

In [12]:

    

lasso


    

    
Out[12]:





Lasso(alpha=1.0, copy_X=True, fit_intercept=True, max_iter=1000,
   normalize=False, positive=False, precompute=False, random_state=None,
   selection='cyclic', tol=0.0001, warm_start=False)

In [13]:

    

# alpha parameter determines how lasso is penalized


    

In [14]:

    

lasso.fit(reg_data, reg_target)


    

    
Out[14]:





Lasso(alpha=1.0, copy_X=True, fit_intercept=True, max_iter=1000,
   normalize=False, positive=False, precompute=False, random_state=None,
   selection='cyclic', tol=0.0001, warm_start=False)

In [15]:

    

np.sum(lasso.coef_ != 0)


    

    
Out[15]:





7

In [16]:

    

lasso_0 = Lasso(0)


    

In [17]:

    

lasso_0.fit(reg_data, reg_target)


    

    




/usr/local/lib/python2.7/site-packages/IPython/kernel/__main__.py:1: UserWarning: With alpha=0, this algorithm does not converge well. You are advised to use the LinearRegression estimator
  if __name__ == '__main__':
/usr/local/lib/python2.7/site-packages/sklearn/linear_model/coordinate_descent.py:432: UserWarning: Coordinate descent with alpha=0 may lead to unexpected results and is discouraged.
  positive)






    
Out[17]:





Lasso(alpha=0, copy_X=True, fit_intercept=True, max_iter=1000,
   normalize=False, positive=False, precompute=False, random_state=None,
   selection='cyclic', tol=0.0001, warm_start=False)

In [18]:

    

np.sum(lasso_0.coef_ != 0)


    

    
Out[18]:





500

In [19]:

    

# how it works:
# Linear Regression minimizes the squared error.
# Lasso still minimizes the squared error but adds a parameter
# that penalizes the error to induce scarcity.


    

In [20]:

    

# Lasso Cross Validation


    

In [21]:

    

from sklearn.linear_model import LassoCV


    

In [22]:

    

lassocv = LassoCV()
lassocv.fit(reg_data, reg_target)


    

    
Out[22]:





LassoCV(alphas=None, copy_X=True, cv=None, eps=0.001, fit_intercept=True,
    max_iter=1000, n_alphas=100, n_jobs=1, normalize=False, positive=False,
    precompute='auto', random_state=None, selection='cyclic', tol=0.0001,
    verbose=False)

In [23]:

    

lassocv.alpha_


    

    
Out[23]:





0.82422898878738993

In [24]:

    

lassocv.coef_[:5]


    

    
Out[24]:





array([-0., -0., -0., -0., -0.])

In [25]:

    

np.sum(lassocv.coef_ != 0)


    

    
Out[25]:





12

In [26]:

    

# you can use this to remove useless features
mask = lassocv.coef_ != 0
new_reg_data = reg_data[:, mask]
new_reg_data.shape


    

    
Out[26]:





(200, 12)

In [ ]: