Simple benchmark to evaluate the impact of compilers on scikit-learn

bench_boosted_trees.py

# Benchmark script for a scikit-learn model that does not use
# BLAS / LAPACK routines
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.datasets import fetch_covtype
from sklearn.cross_validation import train_test_split
from time import time

seed = 0

print("Getting covertype data...")
covtype = fetch_covtype()

# Select a small sub-set of the data as training set to make the
# benchmark fast enough to run
X_train, X_test, y_train, y_test = train_test_split(
    covtype.data, covtype.target, train_size=int(1e4), random_state=seed)

model = GradientBoostingClassifier(n_estimators=100, random_state=seed)

print("Fitting boosted trees on %d samples..."
      % X_train.shape[0])
t0 = time()
model.fit(X_train, y_train)
print("done in %0.3fs" % (time() - t0))

print("Predicting with boosted trees on %d samples..."
      % X_test.shape[0])
t0 = time()
accuracy = model.score(X_test, y_test)
print("done in %0.3fs" % (time() - t0))
print("classification accuracy: %0.3f" % accuracy)

bench_logistic_regression.py

# Benchmark script for a scikit-learn model that does use
# BLAS routines intensively via numpy and scipy
from sklearn.linear_model import LogisticRegressionCV
from sklearn.datasets import fetch_covtype
from sklearn.cross_validation import train_test_split
from time import time
 
seed = 0
 
print("Getting covertype data...")
covtype = fetch_covtype()

# Select a small sub-set of the data as training set to make the
# benchmark fast enough to run
X_train, X_test, y_train, y_test = train_test_split(
    covtype.data, covtype.target, train_size=int(1e4), random_state=seed)
 
model = LogisticRegressionCV(solver='lbfgs')
 
print("Fitting warm-started LR models on %d samples..."
      % X_train.shape[0])
t0 = time()
model.fit(X_train, y_train)
print("done in %0.3fs" % (time() - t0))
 
print("Predicting with linear model on %d samples..."
      % X_test.shape[0])
t0 = time()
accuracy = model.score(X_test, y_test)
print("done in %0.3fs" % (time() - t0))
print("classification accuracy: %0.3f" % accuracy)