start = time.time()
y_pred = clf.predict(X_test)
return np.mean(y_test == y_pred), time.time() - start
try:
dataset = sys.argv[1]
except:
dataset = "usps0"
try:
kernel = sys.argv[2]
except:
kernel = "rbf"
try:
X_train, y_train, X_test, y_test = load_dataset(dataset)
except KeyError:
raise ValueError("Wrong dataset name!")
Nu = np.linspace(0.01, 0.15, 10)
res = [fit_nusvc(X_train, y_train, nu=nu, kernel=kernel) for nu in Nu]
clfs, train_times = zip(*res)
res = [predict(clf, X_test, y_test) for clf in clfs]
accuracies, test_times = zip(*res)
n_samples = X_train.shape[0]
pl.figure()
pl.plot(Nu, [np.mean(clf.n_support_) for clf in clfs])
pl.xlabel("nu")
def parse_kmp(n_nonzero_coefs=200,
n_components=0.5,
metric="rbf",
gamma=0.1,
degree=4,
coef0=1.0,
epsilon=0.0,
n_validate=5,
n_refit=5,
scale=False,
scale_y=False,
check_duplicates=False,
force_cv=False,
cvtype="random",
bars=False,
savefig=""):
op = OptionParser()
op.add_option("--n_folds", action="store", default=5,
dest="n_folds", type="int")
op.add_option("-n", action="store", default=n_nonzero_coefs,
dest="n_nonzero_coefs", type="float")
op.add_option("--n_components", action="store", default=n_components,
dest="n_components", type="float")
op.add_option("--metric", action="store", default=metric, dest="metric",
type="str")
op.add_option("--gamma", action="store", default=gamma, dest="gamma",
type="float")
op.add_option("--degree", action="store", default=degree, dest="degree",
type="int")
op.add_option("--coef0", action="store", default=coef0, dest="coef0",
type="float")
op.add_option("--epsilon", action="store", default=epsilon, dest="epsilon",
type="float")
op.add_option("--n_validate", action="store", default=n_validate,
dest="n_validate", type="int")
op.add_option("--n_refit", action="store", default=n_refit, dest="n_refit",
type="int")
op.add_option("--scale", action="store_true", default=scale, dest="scale")
op.add_option("--scale_y", action="store_true", default=scale_y,
dest="scale_y")
op.add_option("--check_duplicates", action="store_true",
default=check_duplicates, dest="check_duplicates")
op.add_option("--regression", action="store_true", default=scale,
dest="regression")
op.add_option("--force_cv", action="store_true", default=force_cv,
dest="force_cv")
op.add_option("--cvtype", action="store", default=cvtype, dest="cvtype",
type="str")
op.add_option("--bars", action="store_true", default=bars,
dest="bars")
op.add_option("--savefig", action="store", default=savefig, dest="savefig",
type="str")
(opts, args) = op.parse_args()
try:
dataset = args[0]
except:
dataset = "usps"
try:
X_train, y_train, X_test, y_test = load_dataset(dataset)
print "X_train", X_train.shape
if X_test is not None: print "X_test", X_test.shape
return X_train, y_train, X_test, y_test, opts, args
except KeyError:
raise ValueError("Wrong dataset name!")
scale=True,
scale_y=opts.scale_y,
check_duplicates=opts.check_duplicates,
n_validate=opts.n_validate,
epsilon=opts.epsilon,
verbose=1,
random_state=random_state,
n_jobs=-1)
clf.fit(K_train, y_train)
return clf
dataset, opts, random_state = parse_kmp()
try:
X_train, y_train, X_test, y_test = load_dataset(dataset,
proportion_train=0.75,
random_state=random_state)
except KeyError:
raise ValueError("Wrong dataset name!")
print "X_train", X_train.shape
print "X_test", X_test.shape
# PCA view
print "Computing PCA..."
pca = RandomizedPCA(n_components=300)
X_train_pca = pca.fit_transform(X_train)
X_test_pca = pca.transform(X_test)
components_pca = select_components(X_train_pca, y_train,
n_components=opts.n_components,
