def test_compare_scikit_learn():
X = np.array([[-1, -1], [-2, -1], [-3, -2], [1, 1], [2, 1], [3, 2]])
pcas = sklPCA(n_components=2)
pcas.fit(X)
pca = PCA(NDDataset(X))
pca.printev(n_pc=2)
assert_array_almost_equal(pca.sv.data, pcas.singular_values_)
assert_array_almost_equal(pca.ev_ratio.data, pcas.explained_variance_ratio_ * 100.)
dataset = NDDataset.read('irdata/nh4y-activation.spg')
X = dataset.data
pcas = sklPCA(n_components=5)
pcas.fit(X)
dataset = X.copy()
pca = PCA(NDDataset(dataset))
pca.printev(n_pc=5)
assert_array_almost_equal(pca.sv.data[:5], pcas.singular_values_[:5], 4)
assert_array_almost_equal(pca.ev_ratio.data[:5], pcas.explained_variance_ratio_[:5] * 100., 4)
def PCA(data, n_components=2, desired_var_explained=None):
if desired_var_explained is not None:
pca = sklPCA(n_components=desired_var_explained)
else:
pca = sklPCA(n_components=n_components)
pca.fit_transform([data[key] for key in data])
# for i in range(len(pca.explained_variance_ratio_)):
return pca
def _doPCA(self, nComps, syntheticSEDs):
"""Do PCA on the synthetic spectra
@param nComps number of components to keep
@param syntheticSEDs library of synthetic spectra
"""
waveLen, smooth_spectra = sedMapper.get_sed_array(
syntheticSEDs, self.minWavelen, self.maxWavelen, self.nWavelen)
self.specPCA = sklPCA(nComps)
self.specPCA.fit(smooth_spectra)
def _doPCA(self, nComps, syntheticSEDs):
"""Do PCA on the synthetic spectra
@param nComps number of components to keep
@param syntheticSEDs library of synthetic spectra
"""
waveLen, smooth_spectra = sedMapper.get_sed_array(syntheticSEDs,
self.minWavelen, self.maxWavelen, self.nWavelen)
self.specPCA = sklPCA(nComps)
self.specPCA.fit(smooth_spectra)
def _doPCA(self, ncomp, spectra):
"""
"""
specPCA = sklPCA(ncomp)
specPCA.fit(spectra)
self.meanSpec = specPCA.mean_
self.eigenspectra = specPCA.components_
self.eigenvalue_coeffs = np.array(specPCA.transform(spectra))
print "Mean spectrum shape:", self.meanSpec.shape
print "Eigenspectra shape:", self.eigenspectra.shape
print "Eigenvalues shape:", self.eigenvalue_coeffs.shape
def test_compare_scikit_learn():
try:
import_optional_dependency("scikit-learn")
except ImportError:
return
from sklearn.decomposition import PCA as sklPCA
X = np.array([[-1, -1], [-2, -1], [-3, -2], [1, 1], [2, 1], [3, 2]])
pcas = sklPCA(n_components=2)
pcas.fit(X)
pca = PCA(NDDataset(X))
pca.printev(n_pc=2)
assert_array_almost_equal(pca._sv.data, pcas.singular_values_)
assert_array_almost_equal(pca.ev_ratio.data,
pcas.explained_variance_ratio_ * 100.0)
dataset = NDDataset.read("irdata/nh4y-activation.spg")
X1 = dataset.copy().data
pcas = sklPCA(n_components=5, svd_solver="full")
pcas.fit(X1)
X2 = NDDataset(dataset.copy())
pca = PCA(X2)
pca.printev(n_pc=5)
assert_array_almost_equal(pca._sv.data[:5], pcas.singular_values_[:5], 4)
assert_array_almost_equal(pca.ev_ratio.data[:5],
pcas.explained_variance_ratio_[:5] * 100.0, 4)
show()
def _doPCA(self, ncomp, spectra):
"""PCA the SED set in array
@param ncomp number of principle components to keep
@param spectra array of SEDs
"""
specPCA = sklPCA(ncomp)
specPCA.fit(spectra)
self.meanSpec = specPCA.mean_
self.eigenspectra = specPCA.components_
self.eigenvalue_coeffs = np.array(specPCA.transform(spectra))
print "Mean spectrum shape:", self.meanSpec.shape
print "Eigenspectra shape:", self.eigenspectra.shape
print "Eigenvalues shape:", self.eigenvalue_coeffs.shape
def _doPCA(self, ncomp, spectra):
"""PCA the SED set in array
@param ncomp number of principle components to keep
@param spectra array of SEDs
"""
specPCA = sklPCA(ncomp)
specPCA.fit(spectra)
self.meanSpec = specPCA.mean_
self.eigenspectra = specPCA.components_
self.eigenvalue_coeffs = np.array(specPCA.transform(spectra))
print "Mean spectrum shape:", self.meanSpec.shape
print "Eigenspectra shape:", self.eigenspectra.shape
print "Eigenvalues shape:", self.eigenvalue_coeffs.shape
graph.annotate(fig, features_label, (.55, .95))
return fig, AX
if __name__ == '__main__':
from datavyz import ge
# LOADING THE DATA
from sklearn.datasets import load_breast_cancer
data = load_breast_cancer()
# PERFORMING PCA
from sklearn.decomposition import PCA as sklPCA
pca = sklPCA(n_components=4)
pca.fit_transform(data['data'])
# PLOT
fig, AX = ge.components_plot(pca.components_)
ge.savefig(fig, 'docs/components-plot.png')
ge.show()
from sklearn.datasets import load_iris
dataset = load_iris()
fig, ax = ge.parallel_plot(
dataset['data'],
SET_OF_LABELS=[
'sepal length\n(cm)', 'sepal width\n(cm)', 'petal length\n(cm)',
'petal width\n(cm)'
],
def PCA(Xtrain, Xtest, n_components=2):
pca = sklPCA(n_components)
Xtrain = pca.fit_transform(Xtrain)
Xtest = pca.transform(Xtest)
return Xtrain, Xtest
features = nb.loc[i, 'ActMod_trqClth'].tolist()
nbLetters = sum(features)
# nb of appearance of symbols for engine speed
features = features + nb.loc[i, 'Epm_nEng'].tolist()
# nb of appearance of symbols for pedal position
features = features + nb.loc[i, 'APP_r'].tolist()
# duration of hoop
features = (np.array(features) / nbLetters).tolist()
features = features + [
listValuesHoops[i].loc[listValuesHoops[i].index[-1], 'time'] -
listValuesHoops[i].loc[listValuesHoops[i].index[0], 'time']
]
featuresMatrix.append(features)
return pd.DataFrame.from_records(featuresMatrix)
from sklearn.decomposition import PCA as sklPCA
from sklearn.preprocessing import MinMaxScaler
from mpl_toolkits.mplot3d import Axes3D
test = getFeatures(listHoops, listValuesHoops)
# We run PCA
sklPCA = sklPCA(n_components=8)
test2 = MinMaxScaler().fit_transform(test)
output = sklPCA.fit_transform(test2)
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
ax.scatter(output[:, 0], output[:, 1], output[:, 2], c='b', marker='o')
plt.show()