def test_fail_array_fit():
pca = KPCA(n_components=2)
with pytest.raises(ValueError):
pca.fit(X1[1])
def test_fail_array_transform():
pca = KPCA(n_components=2)
pca.fit(X1)
with pytest.raises(ValueError):
pca.transform(X1[1])
def test_proj():
pca = KPCA(n_components=2)
pca.fit(X1[:2])
exp = np.array([[-0.71, -0.71], [0.71, -0.71]])
assert_almost_equal(pca.X_projected_, exp, decimal=2)
def test_reproj_2():
pca = KPCA(n_components=2)
pca.fit(X1)
exp = pca.transform(X1[1, None])
assert_almost_equal(pca.X_projected_[1, None], exp, decimal=2)
def test_default_2components():
pca = KPCA(n_components=2)
pca.fit(X1)
assert pca.X_projected_.shape == (X1.shape[0], 2)
def test_default_0components():
with pytest.raises(AttributeError):
pca = KPCA(n_components=0)
pca.fit(X1)
#plt.show()
plt.savefig('../figs/tutorial/mlxtendex1_2.png')
plt.close()
#This shows linear PCA unable to generate subspace suitable to linearly separate data.
#PCA is unsupevised method, so input data unlabeled.
# Radial base function (RBF) kernel PCA (KPCA)
from mlxtend.data import iris_data
from mlxtend.preprocessing import standardize
from mlxtend.feature_extraction import RBFKernelPCA as KPCA
#Specify 2-component PCA, gamma choice dependent on dataset, obtained via hyperparameter methods such as Grid search.
#Gamma here is gamma said to give 'good' results by creator of tutorial.
kpca = KPCA(gamma=15.0, n_components=2)
#Fit X with above KPA specifications
kpca.fit(X)
#Project X values onto 'new' (higher dimensional?) feature space (rep by 'g' in associated notes)
X_kpca = kpca.X_projected_
#Plot moons but with kernel-projected X
plt.scatter(X_kpca[y==0, 0], X_kpca[y==0, 1],
color='red', marker='o', alpha=0.5)
plt.scatter(X_kpca[y==1, 0], X_kpca[y==1, 1],
color='blue', marker='^', alpha=0.5)
plt.title('First 2 principal components after RBF Kernel PCA')
plt.xlabel('PC1')
plt.ylabel('PC2')
#plt.show()
def test_default_components():
pca = KPCA()
pca.fit(X1)
assert pca.X_projected_.shape == X1.shape
def test_fail_array_transform():
pca = KPCA(n_components=2)
pca.fit(X1)
exp = pca.transform(X1[1])
def test_fail_array_fit():
pca = KPCA(n_components=2)
pca.fit(X1[1])
def test_default_components():
pca = KPCA(n_components=0)
pca.fit(X1)
from sklearn.model_selection import cross_val_score et = ExtraTreesClassifier(n_estimators=300, max_depth=None, random_state=0,verbose=5) scores = cross_val_score(et, X, y,scoring='f1_micro',cv=5,verbose=5) print scores.mean() #32% ,max depth=none, n_est=300 #kernel PCA from mlxtend.feature_extraction import PrincipalComponentAnalysis as PCA from sklearn.ensemble import ExtraTreesClassifier from sklearn.model_selection import cross_val_score pca = PCA(n_components=700) X_pca = pca.fit(X).transform(X) et = ExtraTreesClassifier(n_estimators=500, max_depth=None, random_state=0,verbose=5) scores = cross_val_score(et, X_pca, y,scoring='f1_micro',cv=5,verbose=5) print scores.mean() from mlxtend.feature_extraction import RBFKernelPCA as KPCA kpca = KPCA(gamma=1.0, n_components=700) kpca.fit(X) X_kpca = kpca.fit(X).transform(X) et = ExtraTreesClassifier(n_estimators=500, max_depth=None, random_state=0,verbose=5) scores = cross_val_score(et, X_pca, y,scoring='f1_micro',cv=5,verbose=5) print scores.mean()
marker='^', #triangle marker
alpha=0.5,
)
plt.title('BMI vs glucose by sex after linear PCA')
plt.ylabel('Serum glucose concentration')
plt.xlabel('BMI')
plt.legend([sex1, sex2], ['Sex 1', 'Sex 2'])
#plt.show()
plt.savefig('../../figs/bivariate/subsetkpca1_3')
plt.close()
'''
#kpca with arbitrary gamma
gamma = 500
kpca = KPCA(gamma=gamma, n_components=2)
#Fit X with above KPA specifications
kpca.fit(X)
#Project X values onto 'new' (higher dimensional?) feature space (rep by 'g' in associated notes)
X_kpca = kpca.X_projected_
#Graph after kpca
#generate graph from matrix
for i in range(len(X)):
if sex[i] == 1:
#Sex 1 glucose v BMI
sex1 = plt.scatter(
X_kpca[i][0], #bmi
X_kpca[i][1], #glucose
color='red',
marker='o', #circle marker