Bench arcene, normal vs random.

arcene.py

from sklearn.linear_model import *
import numpy as np 
import matplotlib.pyplot as plt 

# This can be got from, https://github.com/scikit-learn/ml-benchmarks/blob/master/benchmarks/data/arcene_train.labels
# and https://github.com/scikit-learn/ml-benchmarks/blob/master/benchmarks/data/arcene_train.data
f = open("../Downloads/arcene/arcene_train.data")
X = np.fromfile(f, dtype=np.float64, sep=' ')
X = X.reshape(-1, 10000)
f = open("../Downloads/arcene/arcene_train.labels")
y = np.fromfile(f, dtype=np.int32, sep=' ')

r_dual_gap = []
n_dual_gap = []

print "Here"
clf = ElasticNetCV(n_jobs=-1, max_iter=5000)
clf.fit(X, y)
n_iter = [500, 1000, 2000, 5000, 10000, 20000, 30000, 40000, 50000]

for max_iter in n_iter:
    clf_random = ElasticNet(alpha=clf.alpha_, l1_ratio=0.5,
	    max_iter=max_iter, tol=0, random_state=42)
    print "................."
    print "%d number of iterations for random" % max_iter
    clf_random.fit(X, y)
    print clf_random.dual_gap_
    r_dual_gap.append(clf_random.dual_gap_.tolist())
    clf_normal = ElasticNet(alpha=clf.alpha_, l1_ratio=0.5,
	    max_iter=max_iter, tol=0)
    print "................."
    print "%d number of iterations for normal" % max_iter
    clf_normal.fit(X, y)
    print clf_normal.dual_gap_
    n_dual_gap.append(clf_normal.dual_gap_.tolist())

plt.title("Bench arcene with %0.15f" % clf.alpha_)
plt.xlabel("No. of iterations ")
plt.ylabel("Log of dual gap ")
random_ = plt.plot(n_iter, np.log(r_dual_gap), "r-")[0]
normal = plt.plot(n_iter, np.log(n_dual_gap), "b-")[0]
plt.legend([random_, normal], ["Random descent", "Normal descent"])
plt.savefig("best_alpha_arcene.png")