def reduceDataset(self,nr=3,method='PCA'):
'''It reduces the dimensionality of a given dataset using different techniques provided by Sklearn library
Methods available:
'PCA'
'FactorAnalysis'
'KPCArbf','KPCApoly'
'KPCAcosine','KPCAsigmoid'
'IPCA'
'FastICADeflation'
'FastICAParallel'
'Isomap'
'LLE'
'LLEmodified'
'LLEltsa'
'''
dataset=self.ModelInputs['Dataset']
#dataset=self.dataset[Model.in_columns]
#dataset=self.dataset[['Humidity','TemperatureF','Sea Level PressureIn','PrecipitationIn','Dew PointF','Value']]
#PCA
if method=='PCA':
sklearn_pca = sklearnPCA(n_components=nr)
reduced = sklearn_pca.fit_transform(dataset)
#Factor Analysis
elif method=='FactorAnalysis':
fa=FactorAnalysis(n_components=nr)
reduced=fa.fit_transform(dataset)
#kernel pca with rbf kernel
elif method=='KPCArbf':
kpca=KernelPCA(nr,kernel='rbf')
reduced=kpca.fit_transform(dataset)
#kernel pca with poly kernel
elif method=='KPCApoly':
kpca=KernelPCA(nr,kernel='poly')
reduced=kpca.fit_transform(dataset)
#kernel pca with cosine kernel
elif method=='KPCAcosine':
kpca=KernelPCA(nr,kernel='cosine')
reduced=kpca.fit_transform(dataset)
#kernel pca with sigmoid kernel
elif method=='KPCAsigmoid':
kpca=KernelPCA(nr,kernel='sigmoid')
reduced=kpca.fit_transform(dataset)
#ICA
elif method=='IPCA':
ipca=IncrementalPCA(nr)
reduced=ipca.fit_transform(dataset)
#Fast ICA
elif method=='FastICAParallel':
fip=FastICA(nr,algorithm='parallel')
reduced=fip.fit_transform(dataset)
elif method=='FastICADeflation':
fid=FastICA(nr,algorithm='deflation')
reduced=fid.fit_transform(dataset)
elif method == 'All':
self.dimensionalityReduction(nr=nr)
return self
self.ModelInputs.update({method:reduced})
self.datasetsAvailable.append(method)
return self
def dimensionalityReduction(self,nr=5):
'''It applies all the dimensionality reduction techniques available in this class:
Techniques available:
'PCA'
'FactorAnalysis'
'KPCArbf','KPCApoly'
'KPCAcosine','KPCAsigmoid'
'IPCA'
'FastICADeflation'
'FastICAParallel'
'Isomap'
'LLE'
'LLEmodified'
'LLEltsa'
'''
dataset=self.ModelInputs['Dataset']
sklearn_pca = sklearnPCA(n_components=nr)
p_components = sklearn_pca.fit_transform(dataset)
fa=FactorAnalysis(n_components=nr)
factors=fa.fit_transform(dataset)
kpca=KernelPCA(nr,kernel='rbf')
rbf=kpca.fit_transform(dataset)
kpca=KernelPCA(nr,kernel='poly')
poly=kpca.fit_transform(dataset)
kpca=KernelPCA(nr,kernel='cosine')
cosine=kpca.fit_transform(dataset)
kpca=KernelPCA(nr,kernel='sigmoid')
sigmoid=kpca.fit_transform(dataset)
ipca=IncrementalPCA(nr)
i_components=ipca.fit_transform(dataset)
fip=FastICA(nr,algorithm='parallel')
fid=FastICA(nr,algorithm='deflation')
ficaD=fip.fit_transform(dataset)
ficaP=fid.fit_transform(dataset)
'''isomap=Isomap(n_components=nr).fit_transform(dataset)
try:
lle1=LocallyLinearEmbedding(n_components=nr).fit_transform(dataset)
except ValueError:
lle1=LocallyLinearEmbedding(n_components=nr,eigen_solver='dense').fit_transform(dataset)
try:
lle2=LocallyLinearEmbedding(n_components=nr,method='modified').fit_transform(dataset)
except ValueError:
lle2=LocallyLinearEmbedding(n_components=nr,method='modified',eigen_solver='dense').fit_transform(dataset)
try:
lle3=LocallyLinearEmbedding(n_components=nr,method='ltsa').fit_transform(dataset)
except ValueError:
lle3=LocallyLinearEmbedding(n_components=nr,method='ltsa',eigen_solver='dense').fit_transform(dataset)'''
values=[p_components,factors,rbf,poly,cosine,sigmoid,i_components,ficaD,ficaP]#,isomap,lle1,lle2,lle3]
keys=['PCA','FactorAnalysis','KPCArbf','KPCApoly','KPCAcosine','KPCAsigmoid','IPCA','FastICADeflation','FastICAParallel']#,'Isomap','LLE','LLEmodified','LLEltsa']
self.ModelInputs.update(dict(zip(keys, values)))
[self.datasetsAvailable.append(key) for key in keys ]
#debug
#dataset=pd.DataFrame(self.ModelInputs['Dataset'])
#dataset['Output']=self.ModelOutput
#self.debug['Dimensionalityreduction']=dataset
###
return self
def ica(self, n_components=None, sources='left'):
"""Return result from independent component analysis.
X = SA + m
Sklearn's FastICA implementation is used.
When sources=left the sources are returned in the first (left) matrix
and the mixing matrix is returned in the second (right) matrix,
corresponding to X = SA.
When sources=right the sources are returned in the second matrix while
the mixing matrix is returned in the first, corresponding to X = AS.
Parameters
----------
n_components : int, optional
Number of ICA components.
sources : left or right, optional
Indicates whether the sources should be the left or right matrix.
Returns
-------
first : Matrix
Estimated source matrix (S) if sources=left.
second : Matrix
Estimated mixing matrix (A) if sources=right.
mean_vector : brede.core.vector.Vector
Estimated mean vector
References
----------
http://scikit-learn.org/stable/modules/decomposition.html#ica
"""
if n_components is None:
min_shape = min(self.shape[0], len(self._eeg_columns))
n_components = int(np.ceil(sqrt(float(min_shape) / 2)))
ica = FastICA(n_components=n_components)
if sources == 'left':
sources = Matrix(ica.fit_transform(
self.ix[:, self._eeg_columns].values),
index=self.index)
mixing_matrix = Matrix(ica.mixing_.T, columns=self._eeg_columns)
mean_vector = Vector(ica.mean_, index=self._eeg_columns)
return sources, mixing_matrix, mean_vector
elif sources == 'right':
sources = Matrix(ica.fit_transform(
self.ix[:, self._eeg_columns].values.T).T,
columns=self._eeg_columns)
mixing_matrix = Matrix(ica.mixing_, index=self.index)
mean_vector = Vector(ica.mean_, index=self.index)
return mixing_matrix, sources, mean_vector
else:
raise ValueError('Wrong argument to "sources"')
def mixing_matrix(data, n_components, display=True):
features, weights, labels = data
ica = FastICA(n_components=n_components)
ica.fit_transform(features)
mixing = ica.mixing_
if display:
f, ax = plt.subplots(figsize=(10, 4))
sns.heatmap(mixing)
plt.title('Signal Mixing Estimated Matrix')
return mixing
def run_ica(data, comp):
ica = FastICA(n_components=comp, whiten=True, max_iter=5000)
data_out=np.zeros((comp,np.shape(data[0,:,0])[0],np.shape(data[0,0,:])[0]))
for i in range(np.shape(data[0,:,0])[0]):
print i
data_out[:,i,:]=np.transpose(ica.fit_transform(np.transpose(data[:,i,:])))
return data_out
def ica_analysis(self, X_train, X_test, y_train, y_test, data_set_name):
scl = RobustScaler()
X_train_scl = scl.fit_transform(X_train)
X_test_scl = scl.transform(X_test)
##
## ICA
##
ica = FastICA(n_components=X_train_scl.shape[1])
X_ica = ica.fit_transform(X_train_scl)
##
## Plots
##
ph = plot_helper()
kurt = kurtosis(X_ica)
print(kurt)
title = 'Kurtosis (FastICA) for ' + data_set_name
name = data_set_name.lower() + '_ica_kurt'
filename = './' + self.out_dir + '/' + name + '.png'
ph.plot_simple_bar(np.arange(1, len(kurt)+1, 1),
kurt,
np.arange(1, len(kurt)+1, 1).astype('str'),
'Feature Index',
'Kurtosis',
title,
filename)
def best_ica_nba(self):
dh = data_helper()
X_train, X_test, y_train, y_test = dh.get_nba_data()
scl = RobustScaler()
X_train_scl = scl.fit_transform(X_train)
X_test_scl = scl.transform(X_test)
ica = FastICA(n_components=X_train_scl.shape[1])
X_train_transformed = ica.fit_transform(X_train_scl, y_train)
X_test_transformed = ica.transform(X_test_scl)
## top 2
kurt = kurtosis(X_train_transformed)
i = kurt.argsort()[::-1]
X_train_transformed_sorted = X_train_transformed[:, i]
X_train_transformed = X_train_transformed_sorted[:,0:2]
kurt = kurtosis(X_test_transformed)
i = kurt.argsort()[::-1]
X_test_transformed_sorted = X_test_transformed[:, i]
X_test_transformed = X_test_transformed_sorted[:,0:2]
# save
filename = './' + self.save_dir + '/nba_ica_x_train.txt'
pd.DataFrame(X_train_transformed).to_csv(filename, header=False, index=False)
filename = './' + self.save_dir + '/nba_ica_x_test.txt'
pd.DataFrame(X_test_transformed).to_csv(filename, header=False, index=False)
filename = './' + self.save_dir + '/nba_ica_y_train.txt'
pd.DataFrame(y_train).to_csv(filename, header=False, index=False)
filename = './' + self.save_dir + '/nba_ica_y_test.txt'
pd.DataFrame(y_test).to_csv(filename, header=False, index=False)
def align(movie_data, options, args, lrh):
print 'pICA(scikit-learn)'
nvoxel = movie_data.shape[0]
nTR = movie_data.shape[1]
nsubjs = movie_data.shape[2]
align_algo = args.align_algo
nfeature = args.nfeature
randseed = args.randseed
if not os.path.exists(options['working_path']):
os.makedirs(options['working_path'])
# zscore the data
bX = np.zeros((nsubjs*nvoxel,nTR))
for m in range(nsubjs):
bX[m*nvoxel:(m+1)*nvoxel,:] = stats.zscore(movie_data[:, :, m].T ,axis=0, ddof=1).T
del movie_data
np.random.seed(randseed)
A = np.mat(np.random.random((nfeature,nfeature)))
ica = FastICA(n_components= nfeature, max_iter=500,w_init=A,random_state=randseed)
St = ica.fit_transform(bX.T)
ES = St.T
bW = ica.mixing_
R = np.zeros((nvoxel,nfeature,nsubjs))
for m in range(nsubjs):
R[:,:,m] = bW[m*nvoxel:(m+1)*nvoxel,:]
niter = 10
# initialization when first time run the algorithm
np.savez_compressed(options['working_path']+align_algo+'_'+lrh+'_'+str(niter)+'.npz',\
R = R, G=ES.T, niter=niter)
return niter
def ica(self, n_components=None):
"""Return result from independent component analysis.
X = SA + m
Sklearn's FastICA implementation is used.
Parameters
----------
n_components : int, optional
Number of ICA components.
Returns
-------
source : Matrix
Estimated source matrix (S)
mixing_matrix : Matrix
Estimated mixing matrix (A)
mean_vector : brede.core.vector.Vector
Estimated mean vector
References
----------
http://scikit-learn.org/stable/modules/decomposition.html#ica
"""
if n_components is None:
n_components = int(np.ceil(np.sqrt(float(min(self.shape)) / 2)))
ica = FastICA(n_components=n_components)
sources = Matrix(ica.fit_transform(self.values), index=self.index)
mixing_matrix = Matrix(ica.mixing_.T, columns=self.columns)
mean_vector = Vector(ica.mean_, index=self.columns)
return sources, mixing_matrix, mean_vector
def test_inverse_transform():
# Test FastICA.inverse_transform
n_features = 10
n_samples = 100
n1, n2 = 5, 10
rng = np.random.RandomState(0)
X = rng.random_sample((n_samples, n_features))
expected = {(True, n1): (n_features, n1),
(True, n2): (n_features, n2),
(False, n1): (n_features, n2),
(False, n2): (n_features, n2)}
for whiten in [True, False]:
for n_components in [n1, n2]:
n_components_ = (n_components if n_components is not None else
X.shape[1])
ica = FastICA(n_components=n_components, random_state=rng,
whiten=whiten)
with warnings.catch_warnings(record=True):
# catch "n_components ignored" warning
Xt = ica.fit_transform(X)
expected_shape = expected[(whiten, n_components_)]
assert_equal(ica.mixing_.shape, expected_shape)
X2 = ica.inverse_transform(Xt)
assert_equal(X.shape, X2.shape)
# reversibility test in non-reduction case
if n_components == X.shape[1]:
assert_array_almost_equal(X, X2)
def test_inverse_transform():
"""Test FastICA.inverse_transform"""
rng = np.random.RandomState(0)
X = rng.random_sample((100, 10))
rng = np.random.RandomState(0)
X = rng.random_sample((100, 10))
n_features = X.shape[1]
expected = {(True, 5): (n_features, 5),
(True, 10): (n_features, 10),
(False, 5): (n_features, 10),
(False, 10): (n_features, 10)}
for whiten in [True, False]:
for n_components in [5, 10]:
ica = FastICA(n_components=n_components, random_state=rng,
whiten=whiten)
Xt = ica.fit_transform(X)
expected_shape = expected[(whiten, n_components)]
assert_equal(ica.mixing_.shape, expected_shape)
X2 = ica.inverse_transform(Xt)
assert_equal(X.shape, X2.shape)
# reversibility test in non-reduction case
if n_components == X.shape[1]:
assert_array_almost_equal(X, X2)
def filter_frames(self, data):
logging.debug("I am starting the old componenty vous")
data = data[0]
print 'The length of the data is'+str(data.shape)
sh = data.shape
newshape = (np.prod(sh[:-1]), sh[-1])
print "The shape of the data is:"+str(data.shape) + str(newshape)
data = np.reshape(data, (newshape))
# data will already be shaped correctly
logging.debug("Making the matrix")
ica = FastICA(n_components=self.parameters['number_of_components'],
algorithm='parallel',
whiten=self.parameters['whiten'],
w_init=self.parameters['w_init'],
random_state=self.parameters['random_state'])
logging.debug("Performing the fit")
data = self.remove_nan_inf(data) #otherwise the fit flags up an error for obvious reasons
# print "I'm here"
S_ = ica.fit_transform(data)
# print "S_Shape is:"+str(S_.shape)
# print "self.images_shape:"+str(self.images_shape)
scores = np.reshape(S_, (self.images_shape))
eigenspectra = ica.components_
logging.debug("mange-tout")
return [scores, eigenspectra]
def getHeartRate(window, lastHR):
# Normalize across the window to have zero-mean and unit variance
mean = np.mean(window, axis=0)
std = np.std(window, axis=0)
normalized = (window - mean) / std
# Separate into three source signals using ICA
ica = FastICA()
srcSig = ica.fit_transform(normalized)
# Find power spectrum
powerSpec = np.abs(np.fft.fft(srcSig, axis=0))**2
freqs = np.fft.fftfreq(WINDOW_SIZE, 1.0 / FPS)
# Find heart rate
maxPwrSrc = np.max(powerSpec, axis=1)
validIdx = np.where((freqs >= MIN_HR_BPM / SEC_PER_MIN) & (freqs <= MAX_HR_BMP / SEC_PER_MIN))
validPwr = maxPwrSrc[validIdx]
validFreqs = freqs[validIdx]
maxPwrIdx = np.argmax(validPwr)
hr = validFreqs[maxPwrIdx]
print hr
#plotSignals(normalized, "Normalized color intensity")
#plotSignals(srcSig, "Source signal strength")
#plotSpectrum(freqs, powerSpec)
return hr
def test_ica(eng):
t = linspace(0, 10, 100)
s1 = sin(t)
s2 = square(sin(2*t))
x = c_[s1, s2, s1+s2]
random.seed(0)
x += 0.001*random.randn(*x.shape)
x = fromarray(x, engine=eng)
def normalize_ICA(s, aT):
a = aT.T
c = a.sum(axis=0)
return s*c, (a/c).T
from sklearn.decomposition import FastICA
ica = FastICA(n_components=2, fun='cube', random_state=0)
s1 = ica.fit_transform(x.toarray())
aT1 = ica.mixing_.T
s1, aT1 = normalize_ICA(s1, aT1)
s2, aT2 = ICA(k=2, svd_method='direct', max_iter=200, seed=0).fit(x)
s2, aT2 = normalize_ICA(s2, aT2)
tol=1e-1
assert allclose_sign_permute(s1, s2, atol=tol)
assert allclose_sign_permute(aT1, aT2, atol=tol)
def ica(tx, ty, rx, ry):
compressor = ICA(whiten=True) # for some people, whiten needs to be off
newtx = compressor.fit_transform(tx)
newrx = compressor.fit_transform(rx)
em(newtx, ty, newrx, ry, add="wICAtr", times=10)
km(newtx, ty, newrx, ry, add="wICAtr", times=10)
nn(newtx, ty, newrx, ry, add="wICAtr")
def __create_image_obser(self, image_observations) :
"""
Creation of a space in which the images will be compared (learning stage).
Firstly PCA is applied in order to reduce the number of features in the
images. Reduction is done so that 99% of measured variance is covered.
After that, ICA is performed on the coefficients calculated by transforming
(reducing) the face images with PCA. From the learned ICA components
basis_images (vectors), original images coefficients and transformation
for new comming images are extracted.
"""
pca = PCA()
pca.fit(image_observations)
sum = 0
components_to_take = 0
for ratio in pca.explained_variance_ratio_:
components_to_take += 1
sum += ratio
if (sum > 0.99):
break
print("PCA reduces the number of dimensions to: " + str(components_to_take))
pca = PCA(whiten=True, n_components=components_to_take)
self.__transformed_images = pca.fit_transform(image_observations)
self.__transformed_images_mean = np.mean(self.__transformed_images, axis=0)
self.__transformed_images -= self.__transformed_images_mean
self.__pca = pca
ica = FastICA(whiten=True, max_iter=100000)
self.__original_images_repres = ica.fit_transform(self.__transformed_images)
self.__basis_images = ica.mixing_.T
self.__transformation = ica.components_
def independent_component(x, y):
clf = FastICA(random_state=1)
transformed = clf.fit_transform(x.reshape(-1, 1))
comp = clf.components_[0, 0]
mm = clf.mixing_[0, 0]
src_max = transformed.max()
src_min = transformed.min()
return [comp, mm, src_max, src_min]
def transform(data, n_components=3):
features, weights, labels = data
start = time()
ica = FastICA(n_components=n_components)
transformed = ica.fit_transform(features)
elapsed = time() - start
df = pd.DataFrame(transformed)
return df, elapsed
def generate_peoples_results_files(self):
self.np_result = np.c_[self.results[0]['blue'], self.results[0]['green'], self.results[0]['red']]
list_number = len(self.results[0]['blue'])
# ICA
ica = FastICA(n_components=3, fun='logcosh', max_iter=2000)
ica_transformed = ica.fit_transform(self.np_result)
component_all = ica_transformed.ravel([1])
component_1 = component_all[:list_number]
component_2 = component_all[list_number:(2 * list_number)]
component_3 = component_all[(2 * list_number):(3 * list_number)]
# butter_smooth
N = 8
Wn = [1.6 / 30, 4.0 / 30]
t = np.linspace(1 / 30, list_number / 30, list_number)
b, a = signal.butter(N, Wn, 'bandpass', analog=False)
filter_1 = signal.filtfilt(b, a, component_1)
filter_2 = signal.filtfilt(b, a, component_2)
filter_3 = signal.filtfilt(b, a, component_3)
lowess_1 = sm.nonparametric.lowess(filter_1, t, frac=10.0 / list_number)
lowess_2 = sm.nonparametric.lowess(filter_2, t, frac=10.0 / list_number)
lowess_3 = sm.nonparametric.lowess(filter_3, t, frac=10.0 / list_number)
smooths = []
smooth_1 = lowess_1[:, 1]
smooth_2 = lowess_2[:, 1]
smooth_3 = lowess_3[:, 1]
smooths.append(smooth_1)
smooths.append(smooth_2)
smooths.append(smooth_3)
# FFT and spectrum
fft_1 = np.fft.fft(smooth_1, 256)
fft_2 = np.fft.fft(smooth_2, 256)
fft_3 = np.fft.fft(smooth_3, 256)
spectrum_1 = list(np.abs(fft_1) ** 2)
spectrum_2 = list(np.abs(fft_2) ** 2)
spectrum_3 = list(np.abs(fft_3) ** 2)
max1 = max(spectrum_1)
max2 = max(spectrum_2)
max3 = max(spectrum_3)
num_spec1 = spectrum_1.index(max(spectrum_1))
if num_spec1 > (list_number / 2):
num_spec1 = 256 - num_spec1
num_spec2 = spectrum_2.index(max(spectrum_2))
if num_spec2 > (list_number / 2):
num_spec2 = 256 - num_spec2
num_spec3 = spectrum_3.index(max(spectrum_3))
if num_spec3 > (list_number / 2):
num_spec3 = 256 - num_spec3
num_spec = [num_spec1, num_spec2, num_spec3]
max_all = [max1, max2, max3]
max_num = max_all.index(max(max_all))
self.heartRate = int(num_spec[max_num] * 1800 / 256) + 1
return smooths[max_num]
def _fit_local(self, data):
from sklearn.decomposition import FastICA
from numpy import random
random.seed(self.seed)
model = FastICA(n_components=self.k, fun="cube", max_iter=self.max_iter, tol=self.tol, random_state=self.seed)
signals = model.fit_transform(data)
return signals, model.mixing_.T
def fit_transform_ica(X):
ica = FastICA(n_components=50, max_iter=2000, tol=0.05, algorithm='parallel', fun='cube', fun_args={'alpha': 1.0}, random_state=42) #26 36 76
start = time.time()
X = ica.fit_transform(X)
end = time.time()
print "Done!\nFit ICA transform time (secs): {:.3f}".format(end - start)
return X, ica
def print_kurtosis(scaled_data):
#print the kurtosis of the scaled data
print "Kurotsis of original DF:", kurtosis(scaled_data)
#print the kurtosis of the ICA transformed columns
for i in range(1,len(scaled_data[0])+1):
ica = FastICA(n_components=i)
ica_fit = ica.fit_transform(scaled_data)
print "Kurtosis of ICA Transformed data when i=" + str(i) + ":", kurtosis(ica_fit)
def ICA(model_data, components = None, transform_data = None):
t0 = time()
ica = FastICA(n_components=components)
if transform_data == None:
projection = ica.fit_transform(model_data)
else:
ica.fit(model_data)
projection = ica.transform(transform_data)
print "ICA Time: %0.3f" % (time() - t0)
return projection
class ICA(Transform):
def __init__(self, dependency, n_components=6):
self.ica = FastICA(n_components)
self.dependency = dependency
def requires(self):
return [self.dependency]
def apply(self, data):
return self.ica.fit_transform(data.T).T
def fun_doICA(X, nc):
'''
Perform ICA and sort signals by bimodality
'''
ica = FastICA(n_components = nc)
Sest = ica.fit_transform(X)
A = ica.mixing_
S = fun_sort_bimod(Sest)
out = {'S':S, 'A':A}
return out
def pca_ica(mov, components=50, batch=1000, mu=0.5, ica_func='logcosh', show_status=True):
"""Perform iterative PCA/ICA ROI extraction
Parameters
----------
mov : pyfluo.Movie
input movie
components : int
number of independent components to return
batch : int
number of pixels to load into memory simultaneously. More leads to a better fit, but requires more memory
mu : float
from 0-1. In spatiotemporal ICA, closer to 1 means more weight on spatial
ica_func : str
cdf for entropy maximization in ICA
show_status : bool
show time elapsed while running
Returns
-------
Array of shape (n,y,x) where n is number of components, and y,x correspond to shape of mov
"""
if show_status:
p = mup.Process(target=display_time_elapsed)
p.start()
eigenseries, eigenframes,_proj = ipca(mov, components, batch)
# normalize the series
frame_scale = mu / np.max(eigenframes)
frame_mean = np.mean(eigenframes, axis = 0)
n_eigenframes = frame_scale * (eigenframes - frame_mean)
series_scale = (1-mu) / np.max(eigenframes)
series_mean = np.mean(eigenseries, axis = 0)
n_eigenseries = series_scale * (eigenseries - series_mean)
# build new features from the space/time data
# and compute ICA on them
eigenstuff = np.concatenate([n_eigenframes, n_eigenseries])
ica = FastICA(n_components=components, fun=ica_func)
joint_ics = ica.fit_transform(eigenstuff)
# extract the independent frames
num_frames, h, w = mov.shape
frame_size = h * w
ind_frames = joint_ics[:frame_size, :]
ind_frames = np.reshape(ind_frames.T, (components, h, w))
if show_status: p.terminate()
return ind_frames
def Get_Result(filename1,filename2,withbeam=True):
'''filename1: 21cm
filename2 :foreground or 21cm+fg+beam
'''
call('rm *.eps', shell=True)
Plot = False
pol = 0
N = 4
# withbeam=True
############read data#####################
map1 = tt.ICA.ReadMap(filename1)
map2 = tt.ICA.ReadMap(filename2)
if withbeam==True:
map = map2
Freq_num = map.shape[0]
S = map.T
del map
else:
map=map1[:,pol]+map2[:,pol]
Freq_num = map.shape[0]
S=map.T
############FastICA#######################
ica = FastICA(
n_components=N,
algorithm='parallel',
whiten=True,
fun='logcosh',
fun_args=None,
max_iter=200,
tol=0.0001,
w_init=None,
random_state=None)
S_ = ica.fit_transform(S)
A_ = ica.mixing_
##########################################
# tt.ICA.GetComponent(N, S_)
# re=tt.ICA.rebuild(N, A_, S_, 0, Plot=False)
# residuals=tt.ICA.RESULT(0,N,pol,A_,S_,map1,map2,map3,Plot)
res = []
# return (map1[100,pol],map1[100,pol]+map2[100,pol]-re)
##########################################
for i in range(Freq_num):
res.append(tt.ICA.RESULT(i, N, pol, A_, S_, map1, map2, Plot, withbeam))
print res[-1]
plt.close('all')
res = np.array(res)
resx = np.linspace(700, 800, Freq_num, endpoint=True)
# plt.plot(resx,res,label='freq_%d pixel_%d'%(F,P))
# plt.show()
return np.c_[resx,res]
def decompose(data, data_cols=None, kind='ICA', n_components=None, iterations=300):
decompositor = None
if kind == 'ICA':
decompositor = FastICA(n_components=n_components, max_iter=iterations)
elif kind == 'PCA':
decompositor = PCA(n_components=n_components)
elif kind == 'Kernel':
decompositor = KernelPCA(n_components=n_components, max_iter=iterations)
transformed_data = decompositor.fit_transform(data.as_matrix(data_cols))
# columns = ['pca{0:0>3}'.format(idx) for idx, value in enumerate(transformed_data, start=0)]
dataframe = pd.DataFrame(transformed_data, index=data.index)
dataframe.insert(len(dataframe.columns), 'class', data['class'])
return dataframe
def calcICA(delta_data, components):
data = preprocess(delta_data)
ica = FastICA(n_components=components)
x_ica = ica.fit_transform(data['cleanMatrix'])
ica_fill = np.ones((delta_data.shape[0],components))*np.nan
ica_fill[data['cleanind']] = x_ica
ica_weights = ica.components_.T
delta_ica = {'transform':ica_fill,
'weights' : ica_weights,
}
return delta_ica
def fast_ica(brain, components):
ica = FastICA(n_components=components)
S_ = ica.fit_transform(brain) # Reconstruct signals
A_ = ica.mixing_ # Get estimated mixing matrix
return S_
# outfile = infile.split('.')[0] + 'fast_ica.csv'
# with open(outfile, 'wb') as s:
# writer = csv.writer(s)
# writer.writerows(S_)
# return outfile
# S2 = np.sin(2 * time) # Signal 1 : sinusoidal signal
S2 = np.sign(np.sin(3 * time)) # Signal 2 : square signal
# S_2 = signal.sawtooth(2 * np.pi * time) # Signal 3: saw tooth signal
# S2 += 0.2 * np.random.normal(size=S2.shape) # Add noise
S = np.c_[S1, S2]
S /= S.std(axis=0) # Standardize data
# Mix data
A = np.c_[np.ones([2, 1]), np.random.rand(2, 2)] # Mixing matrix
X = np.dot(S, A) # Generate observations
# Compute ICA
ica = FastICA(n_components=2)
S_ica = ica.fit_transform(X)
# #############################################################################
# Plot results
plt.figure()
# plot source signal
ax = plt.subplot(4, 1, 1)
ax.set_title('Source 1')
ax.plot(S1)
ax = plt.subplot(4, 1, 2)
ax.set_title('Source 1')
ax.plot(S2)
# plot mixing signal
ax = plt.subplot(4, 1, 3)
ax.set_title('Observations')
S = np.c_[s1, s2, s3]
S += 0.2 * np.random.normal(size=S.shape) # Add noise
S /= S.std(axis=0) # Standardize data
# Mix data
A = np.array([[1, 1, 1], [0.5, 2, 1.0], [1.5, 1.0, 2.0]]) # Mixing matrix
X = np.dot(S, A.T) # Generate observations
###############################################################################
# Now try to recover the sources
# ------------------------------
# compute ICA
ica = FastICA(n_components=3)
S_ = ica.fit_transform(X) # Get the estimated sources
A_ = ica.mixing_ # Get estimated mixing matrix
# compute PCA
pca = PCA(n_components=3)
H = pca.fit_transform(X) # estimate PCA sources
plt.figure(figsize=(9, 6))
models = [X, S, S_, H]
names = [
'Observations (mixed signal)', 'True Sources', 'ICA estimated sources',
'PCA estimated sources'
]
colors = ['red', 'steelblue', 'orange']
def apply_ICA(proj_data, proj_weights=None):
ica = FastICA(n_components=2, random_state=RANDOM_SEED);
result = ica.fit_transform(proj_data.copy().T); # Copy needed because ICA whitens the input matrix
return result;
def vis_embeddings(dim_red_method, epochs, sample):
n_comp = 2
x_train = epochs.get_data()
x_train = x_train.transpose(0, 2, 1).reshape(-1, x_train.shape[1])
x_train = StandardScaler().fit_transform(x_train)
y_train = get_y_train(sample)
inds = np.arange(15, 8000, 50)
x_train = x_train[inds]
y_train = y_train[inds]
print('fitting {}'.format(dim_red_method))
if dim_red_method == 'pca':
pca = PCA(n_components=n_comp)
reduced_data = pca.fit_transform(x_train)
elif dim_red_method == 'ica':
ica = FastICA(n_components=n_comp)
reduced_data = ica.fit_transform(x_train)
elif dim_red_method == 'se':
se = SpectralEmbedding(n_components=n_comp)
reduced_data = se.fit_transform(x_train)
elif dim_red_method == 'tsne':
pca = PCA(n_components=50)
pca_data = pca.fit_transform(x_train)
tsne = TSNE(n_components=n_comp,
verbose=1,
perplexity=10,
learning_rate=200)
reduced_data = tsne.fit_transform(pca_data)
else:
raise ValueError("{} method not implemented".format(dim_red_method))
print('fitting done')
if n_comp == 2:
reduced_data_df = pd.DataFrame(data=reduced_data,
columns=['PC1', 'PC2'])
elif n_comp == 3:
reduced_data_df = pd.DataFrame(data=reduced_data,
columns=['PC1', 'PC2', 'PC3'])
y_train_df = pd.DataFrame(data=y_train, columns=["labels"])
final_df = pd.concat([reduced_data_df, y_train_df[['labels']]], axis=1)
if n_comp == 2:
sns.set()
palette = sns.color_palette("bright", 8)
ax = sns.scatterplot(x='PC1',
y='PC2',
hue='labels',
data=final_df,
palette=palette,
legend='full')
ax.set(xlabel='PC1',
ylabel='PC2',
title='2 component {}'.format(dim_red_method))
plt.show()
elif n_comp == 3:
ax = plt.figure(figsize=(16, 10)).gca(projection='3d')
ax.scatter(xs=final_df["PC1"],
ys=final_df["PC2"],
zs=final_df["PC2"],
c=final_df["labels"],
cmap='tab10')
ax.set_xlabel('PC1')
ax.set_ylabel('PC2')
ax.set_zlabel('PC3')
plt.show()
s3 = signal.sawtooth(2 * np.pi * time) # Signal 3: saw tooth signal
S = np.c_[s1, s2, s3]
S += 0.2 * np.random.normal(size=S.shape) # Add noise
S /= S.std(axis=0) # Standardize data
# Mix data
A = np.array([[1, 1, 1], [0.5, 2, 1.0], [1.5, 1.0, 2.0]])
B = np.array([[1, 3, 2], [2, 0.1, 1.0], [2.0, 1.5, 0.5]])
C = np.array([[1, 2, 0.5], [2, 0.5, 0.5], [0.5, 0.5, 2.0]])
# Mixing matrix
ica = FastICA(n_components=3)
XA = np.dot(S, A.T) # Generate observations
SA_ = ica.fit_transform(XA) # Reconstruct signals
A_ = ica.mixing_ # Get estimated mixing matrix
XB = np.dot(S, B.T) # Generate observations
SB_ = ica.fit_transform(XB) # Reconstruct signals
B_ = ica.mixing_ # Get estimated mixing matrix
XC = np.dot(S, C.T) # Generate observations
SC_ = ica.fit_transform(XC) # Reconstruct signals
C_ = ica.mixing_ # Get estimated mixing matrix
plt.figure()
models = [XA, XB, XC, S, SA_, SB_, SC_]
names = [
'Observations(mixed signal by A)', 'Observations(mixed signal by B)',
'Observations(mixed signal by C)', 'True Sources',
def run_and_fit(self,
model_string,
nr_components,
nr_timepoints,
nr_neurons,
lambd=0):
np.random.seed(7)
#X=self.simulate_data(nr_components,nr_timepoints,nr_neurons)
X = self.simulate_data_w_noise(nr_components,
nr_timepoints,
nr_neurons,
noise_ampl_mult=4)
if model_string == 'EnsemblePursuit':
options_dict = {'seed_neuron_av_nr': 10, 'min_assembly_size': 1}
ep_pt = EnsemblePursuitPyTorch(n_ensembles=nr_components,
lambd=lambd,
options_dict=options_dict)
U, V = ep_pt.fit_transform(X)
self.U = U.numpy()
self.V = V.numpy().T
if model_string == 'EnsemblePursuitNumpy':
options_dict = {'seed_neuron_av_nr': 10, 'min_assembly_size': 1}
ep_np = EnsemblePursuitNumpy(n_ensembles=nr_components,
lambd=lambd,
options_dict=options_dict)
U, V, self.corrs = ep_np.fit_transform(X)
self.U = U
self.V = V.T
if model_string == 'ICA':
ica = FastICA(n_components=nr_components, random_state=7)
self.V = ica.fit_transform(X.T).T
self.U = ica.mixing_
if model_string == 'PCA':
pca = PCA(n_components=nr_components, random_state=7)
self.V = pca.fit_transform(X.T).T
self.U = pca.components_.T
if model_string == 'sparsePCA':
spca = SparsePCA(n_components=nr_components, random_state=7)
self.V = spca.fit_transform(X.T).T
self.U = spca.components_.T
if model_string == 'NMF':
X -= X.min(axis=0)
nmf = NMF(n_components=nr_components,
init='nndsvd',
random_state=7,
alpha=lambd,
l1_ratio=0.5)
self.V = nmf.fit_transform(X.T).T
self.U = nmf.components_.T
if model_string == 'LDA':
X -= X.min(axis=0)
nmf = LatentDirichletAllocation(n_components=nr_components,
random_state=7)
self.V = nmf.fit_transform(X.T).T
self.U = nmf.components_.T
print('SHPS', self.U.shape, self.V.shape)
self.orig = X
self.approx = self.U @ self.V
print('orig', self.orig.shape)
print('approx', self.approx.shape)
def add_pld_params(model_params,
fluxes,
pld_intensities,
n_pld=9,
order=3,
add_unity=True,
do_pca=True,
do_ica=False,
do_std=True,
pca_cut=False,
n_ppm=1.0,
start_unity=False,
verbose=False):
# Make a local copy
pld_intensities = pld_intensities.copy()
if len(pld_intensities) != n_pld * order:
pld_intensities = np.vstack(
[list(pld_intensities**k) for k in range(1, order + 1)])
# check that the second set is the square of the first set, and so onself.
for k in range(order):
assert (np.allclose(pld_intensities[:n_pld]**(k + 1),
pld_intensities[k * n_pld:(k + 1) * n_pld]))
if do_pca or do_ica: do_std = True
stdscaler = StandardScaler()
pld_intensities = stdscaler.fit_transform(
pld_intensities.T) if do_std else pld_intensities.T
if do_pca:
pca = PCA()
pld_intensities = pca.fit_transform(pld_intensities)
evrc = pca.explained_variance_ratio_.cumsum()
n_pca = np.where(evrc > 1.0 - n_ppm / ppm)[0].min()
if pca_cut: pld_intensities = pld_intensities[:, :n_pca]
if verbose: print(evrc, n_pca)
if do_ica:
ica = FastICA()
pld_intensities = ica.fit_transform(pld_intensities)
# evrc = ica.explained_variance_ratio_.cumsum()
# n_ica = np.where(evrc > 1.0-n_ppm/ppm)[0].min()
# if ica_cut: pld_intensities = pld_intensities[:,:n_ica]
#
# if verbose: print(evrc, n_ica)
if add_unity:
pld_intensities = np.vstack(
[pld_intensities.T,
np.ones(pld_intensities.shape[0])]).T
pld_coeffs = np.linalg.lstsq(
pld_intensities, fluxes)[0] if not start_unity else np.ones(
pld_intensities.shape[1]) / pld_intensities.shape[1.0]
n_pld_out = n_pca if do_pca and pca_cut else n_pld * order
for k in range(n_pld_out):
model_params.add_many(('pld{}'.format(k), pld_coeffs[k], True))
# if add_unity: model_params.add_many(('pld{}'.format(n_pld_out), pld_coeffs[n_pld_out], True)) # FINDME: Maybe make min,max = 0,2 or = 0.9,1.1
if add_unity:
model_params.add_many(
('pldBase', pld_coeffs[n_pld_out],
True)) # FINDME: Maybe make min,max = 0,2 or = 0.9,1.1
if verbose:
[
print('{:5}: {}'.format(val.name, val.value))
for val in model_params.values() if 'pld' in val.name.lower()
]
return model_params, pld_intensities.T
s1 = np.sin(2 * time) # Signal 1 : sinusoidal signal s2 = np.sign(np.sin(3 * time)) # Signal 2 : square signal s3 = signal.sawtooth(2 * np.pi * time) # Signal 3: saw tooth signal S = np.c_[s1, s2, s3] S += 0.2 * np.random.normal(size=S.shape) # Add noise S /= S.std(axis=0) # Standardize data # Mix data A = np.array([[1, 1, 1], [0.5, 2, 1.0], [1.5, 1.0, 2.0]]) # Mixing matrix X = np.dot(S, A.T) # Generate observations # Compute ICA ica = FastICA(n_components=3) S_ = ica.fit_transform(X) # Reconstruct signals A_ = ica.mixing_ # Get estimated mixing matrix # We can `prove` that the ICA model applies by reverting the unmixing. assert np.allclose(X, np.dot(S_, A_.T) + ica.mean_) # For comparison, compute PCA pca = PCA(n_components=3) H = pca.fit_transform(X) # Reconstruct signals based on orthogonal components ############################################################################### # Plot results plt.figure() models = [X, S, S_, H]
gs.fit(labels_EM_PCA.reshape(-1, 1), dataY)
tmp = pd.DataFrame(gs.cv_results_)
tmp.to_csv(out + 'QSAR NN EM PCA.csv')
best_indices = tmp.index[tmp['rank_test_score'] == 1].tolist()
best_em = best_em.append(
{
'Layers': str(tmp.iloc[best_indices[0], 4]),
'Iterations': tmp.iloc[best_indices[0], 5],
'Score': tmp.iloc[best_indices[0], 12]
},
ignore_index=True)
# Fit/transform with FastICA
print("Running FastICA...")
ica = FastICA(n_components=10, random_state=5)
dataX_ICA = ica.fit_transform(dataX)
# Run KM
print("Running k-means...")
model = KMeans(n_clusters=km)
labels_KM_PCA = model.fit_predict(dataX_ICA)
grid = {
'NN__hidden_layer_sizes': nn_arch,
'NN__max_iter': nn_iter,
'NN__learning_rate_init': [0.016],
'NN__alpha': [0.316227766]
}
mlp = MLPClassifier(activation='relu',
early_stopping=True,
random_state=5)
def ica(self, whiten = True):
ica = FastICA(n_components = 5, whiten = whiten)
self.train = ica.fit_transform(self.train)
print("Transforming...")
n_comp = 50
# tSVD
tsvd = TruncatedSVD(n_components=n_comp, random_state=420)
tsvd_results_train = tsvd.fit_transform(train_df)
tsvd_results_test = tsvd.transform(test_df)
# PCA
pca = PCA(n_components=n_comp, random_state=420)
pca2_results_train = pca.fit_transform(train_df)
pca2_results_test = pca.transform(test_df)
# ICA
ica = FastICA(n_components=n_comp, random_state=420)
ica2_results_train = ica.fit_transform(train_df)
ica2_results_test = ica.transform(test_df)
# GRP
grp = GaussianRandomProjection(n_components=n_comp, eps=0.1, random_state=420)
grp_results_train = grp.fit_transform(train_df)
grp_results_test = grp.transform(test_df)
# SRP
srp = SparseRandomProjection(n_components=n_comp,
dense_output=True,
random_state=420)
srp_results_train = srp.fit_transform(train_df)
srp_results_test = srp.transform(test_df)
############
return n_components_ica
runs = (("data/creditcards_train.arff", "Credit Default", "d1"),
("data/htru_train.arff", "Pulsar Detection", "d2"))
for (fname, label, abbrev) in runs:
X, y, feature_names = load_data(fname)
# model selection (optimal number of components)
n_components = optimize_components(X, feature_names, label, abbrev)
# save as new set of features
ica = FastICA(n_components=n_components, random_state=SEED)
start_time = time.perf_counter()
df = pd.DataFrame(ica.fit_transform(X))
run_time = time.perf_counter() - start_time
print(label + ": run time = " + str(run_time))
print(label + ": iterations until convergence = " + str(ica.n_iter_))
df.to_pickle(path.join(PKL_DIR, abbrev + "_ica.pickle"))
# parallel coordinates plot
visualizer = ParallelCoordinates(sample=0.2, shuffle=True, fast=True)
visualizer.fit_transform(df, y)
visualizer.ax.set_xticklabels(visualizer.ax.get_xticklabels(),
rotation=45,
horizontalalignment='right')
visualizer.finalize()
plt.savefig(path.join(PLOT_DIR, abbrev + "_ica_parallel.png"),
bbox_inches='tight')
visualizer.show()
# shape
print('Shape train: {}\nShape test: {}'.format(train.shape, test.shape))
y_train = train["y"]
y_mean = np.mean(y_train)
#PCA/ICA for dimensionality reduction
n_comp = 10
# PCA
pca = PCA(n_components=n_comp, random_state=42)
pca2_results_train = pca.fit_transform(train.drop(["y"], axis=1))
pca2_results_test = pca.transform(test)
# ICA
ica = FastICA(n_components=n_comp, random_state=42)
ica2_results_train = ica.fit_transform(train.drop(["y"], axis=1))
ica2_results_test = ica.transform(test)
train_cols = [col for col in list(train)]
test_cols = [col for col in list(test)]
print(train_cols)
print(test_cols)
train.drop(train_cols, axis=1, inplace=True)
test.drop(test_cols, axis=1, inplace=True)
# Append decomposition components to datasets
for i in range(1, n_comp + 1):
train['pca_' + str(i)] = pca2_results_train[:, i - 1]
test['pca_' + str(i)] = pca2_results_test[:, i - 1]
plt.vlines(0, min_y, max_y, linewidth=2)
plt.xlim(min_x, max_x)
plt.ylim(min_y, max_y)
plt.title(title)
pml.savefig(f'{file_name}.pdf')
plt.show()
np.random.seed(2)
N = 100
A = np.array([[2, 3], [2, 1]]) * 0.3 # Mixing matrix
S_uni = (np.random.rand(N, 2) * 2 - 1) * np.sqrt(3)
X_uni = S_uni @ A.T
pca = PCA(whiten=True)
S_pca = pca.fit(X_uni).transform(X_uni)
ica = FastICA()
S_ica = ica.fit_transform(X_uni)
S_ica /= S_ica.std(axis=0)
plot_samples(S_uni, 'Uniform Data', 'ica-uniform-source')
plot_samples(X_uni, 'Uniform Data after Linear Mixing', 'ica-uniform-mixed')
plot_samples(S_pca, 'PCA Applied to Mixed Data from Uniform Source',
'ica-uniform-PCA')
plot_samples(S_ica, 'ICA Applied to Mixed Data from Uniform Source',
'ica-uniform-ICA')
label='mean_kurtosis')
line2, = plt.plot(k_arr,
kurt_var,
color='b',
marker='o',
label='variance of kurtosis')
plt.legend(handler_map={line1: HandlerLine2D(numpoints=2)})
plt.ylabel(' kurtosis')
plt.xlabel('Number of components')
plt.show()
return None
kurt(X, y, 20)
ica = FastICA(n_components=11, random_state=0)
ica_2d = ica.fit_transform(X)
X_ica = ica.transform(X)
plt.scatter(ica_2d[:, 0],
ica_2d[:, 1],
c=y,
cmap="RdGy",
edgecolor="None",
alpha=1,
vmin=75,
vmax=150)
plt.colorbar()
plt.title('ICA Scatter Plot')
def plot_samples(S, axis_list=None):
plt.scatter(S[:, 0],
x = dataset.iloc[:, 1:-5]
y = dataset.iloc[:, -5:]
#collect the wavelengths in the spectra
wavelengths = dataset.columns[1:-5]
#list of constituents
constituents = list(dataset.columns[-5:])
from sklearn.decomposition import FastICA
from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score
from math import sqrt
from numpy import savetxt
#find out the ICs on the whole dataset
transformer = FastICA(n_components=n_IC, random_state=rdnSeed, max_iter=4000)
IC = transformer.fit_transform(x.T)
M = transformer.mixing_
#set the target variable
target = constituents[
constituentIndex] #change the index to fit the model to different constituents
y = np.array(y[target])
#create the dataframe so that we can drop the missing values both in x and y
df = np.append(M, y.reshape(len(y), 1), axis=1)
columns_ = np.append(wavelengths, [target]).transpose()
df = pd.DataFrame(data=df)
#drop the rows that has missing values in any columns and reset the index
df.dropna(inplace=True)
c = 0
X_inner_trainingset = X_all[X_arr[subtrain[train], ].reshape(
(numDay - 2) * duration), ]
Y_inner_trainingset = Y_arr[subtrain[train], ].reshape(
(numDay - 2) * duration)
X_validate = X_all[X_arr[subtrain[validate], ], ].reshape(
duration, 600)
Y_validate = Y_all[X_arr[subtrain[validate]]].reshape(duration)
for C in components:
print(test, subtrain[validate], subtrain[train], C)
ica = FastICA(n_components=C, max_iter=5000,
tol=0.0001) #tol = 0.001
X_inner_train = ica.fit_transform(
X_inner_trainingset
) #pull components from ica fit transformation
X_inner_test = ica.transform(X_validate)
clf = svm.SVC(kernel='linear',
class_weight='balanced',
probability=True)
y_inner_score = clf.fit(
X_inner_train,
Y_inner_trainingset).decision_function(X_inner_test)
fpr, tpr, _ = roc_curve(Y_validate, y_inner_score)
roc_auc[t, v, c] = auc(fpr, tpr)
c += 1
v += 1
def transform(self, graph_file, first_node=None):
logging.info('loading graph')
"""
input: csv file of graph; formate: start_node, end_node, weight
output: graph, a list, the elements are tuples, like [(1, 2, 1) (3, 1, 1) (2, 3, 1)]
count amount of nodes from G
"""
self.graph = self.load_graph(graph_file) # obtain a array of graph
self.node_count = self.find_node_count(
self.graph) # find the number of nodes in graph
self.edge_count = len(self.graph)
print("nodes:", self.node_count)
print("edges:", self.edge_count)
self.node_range = range(1, self.node_count + 1)
logging.info('computing distance matrix')
self.distance_matrix = self.compute_distance_matrix(
self.graph, self.node_count)
# self.distance_matrix = self.nomalization_distance_mtrix(distance_matrix=self.distance_matrix) # nomalized distance matrix
############################## adjacency matrix ##########################################
self.adjacency_matrix = self.get_adjacency_matrix(
self.graph, self.node_count)
###########################################################################
if first_node is None:
"""self.first_node = randint(0, self.node_count) + 1 # Choose the first pivot from V randomly."""
self.first_node = randint(1, self.node_count)
else:
self.first_node = first_node # Specify the first pivot.
logging.info('finding pivots')
"""
dimensions=m
choose m pivots according to k-center.
"""
#####################################################
if self.pivot_select == "randomly":
self.pivot_nodes = self.choose_pivots_randomly(
dimension=self.dimension, number_nodes=self.node_count)
#####################################################
else:
self.pivot_nodes = self.choose_pivot_points(
self.graph, self.dimension) # self.pivot_nodes: a list
logging.info('drawing graph in high dimensional space')
"""
note that the number of pivot nodes is the same as dimension.
formate of points:
G=(V, E)
|V|=n, dimensions = m = pivots
d(vi, pj) denotes a distance computered by Dijkstra's algorithm in a G.
p1 p2 p3 ... pm
v1 d(v1, p1) d(v1, p2) d(v1, p3) d(v1, pm)
v2 .
v3 .
v4 . .
. . .
. . .
. .
vn d(vn, p1) ... d(vn, pm)
"""
self.points = list(
map(
lambda i: tuple(self.distance_matrix[i - 1, p - 1]
for p in self.pivot_nodes), self.node_range))
if self.normalization is True:
##############################################################################################################
self.points = self.nomalization_distance_mtrix(
distance_matrix=self.points) # nomalized self.points
##############################################################################################################
logging.info('project into a low dimension use PCA')
if self.version == "HDE-SV":
if self.dimension == 2:
self.transformed_points = np.array(self.points)
"""
PCA:
input array-like: shape of self.points = (n_sample, n_feature)
output array-like: shape of self.transformed_points = (n_sample, n_component)
"""
if self.version == "HDE": # PCA denotes that algorithm uses PCA to decomposite original space.
pca = PCA(n_components=2, copy=True)
self.transformed_points = pca.fit_transform(self.points)
if self.version == "HDE-Level": # PCA denotes that algorithm uses PCA to decomposite original space.
pca = PCA(n_components=3, copy=True)
self.transformed_points = pca.fit_transform(self.points)
pca = PCA(n_components=2, copy=True)
self.transformed_points = pca.fit_transform(
self.transformed_points)
'''
replace initial version as paper. by mty 2017-8-9
'''
if self.version == "HDE-PIT": # PIT denotes that algorithm uses poweriteration to computer eigenvectors for decomposition space.
X, S = self.covariance(self.points)
# X = np.array(self.points).T
# X = X.astype(float)
U = self.poweriteration(S, epsilon=self.epsilon)
self.transformed_points = self.decomposition_space(X, U)
if self.node_count == (self.edge_count +
1): # determine wether it is a tree.
FR = FR_Algorithm(number_of_nodes=self.node_count,
initial_temperature=self.initial_temperature,
cooling_factor=self.cooling_factor,
factor_attract=self.factor_attract,
factor_repulsion=self.factor_repulsion)
# use FR to fine-tune
self.transformed_points = FR.apply_force_directed_algorithm(
iteration=self.fr_iteration,
graph=self.graph,
coord_decomposition=self.transformed_points)
if self.version == "HDE-MDS": # HDE-MDS denotes that algorithm combines with MDS.
hde_mds = MDS() # MDS object
self.transformed_points = hde_mds.fit_transform(self.points)
if self.version == "Pivot-MDS": # Pivot-MDS denotes that original version of Pivot MDS.
pivot_mds = PivotMDS(d=self.distance_matrix,
pivots=self.dimension) # PivotMDS object
self.transformed_points = pivot_mds.optimize()
if self.version == "HDE-FICA": # FICA denotes that algorithm uses Fast ICA to decomposite original space.
# fun, Could be either 'logcosh', 'exp', or 'cube'.
fica = FastICA(n_components=2)
# print(np.array(self.points).shape)
self.transformed_points = fica.fit_transform(self.points)
# print(np.array(self.transformed_points).shape)
# FR = FR_Algorithm(number_of_nodes=self.node_count, initial_temperature=self.initial_temperature,
# cooling_factor=self.cooling_factor, factor_attract=self.factor_attract, factor_repulsion=self.factor_repulsion)
# # use FR to fine-tune
# self.transformed_points = FR.apply_force_directed_algorithm(iteration=self.fr_iteration, graph=self.graph, coord_decomposition=self.transformed_points)
if self.version == "HDE-KPCA": # FPCA denotes that algorithm uses kernel PCA to decomposite original space.
kpca = KernelPCA(n_components=2,
kernel=self.kpca_fun,
gamma=self.gamma)
self.transformed_points = kpca.fit_transform(self.points)
if self.version == "HDE-NMF":
nmf = NMF(n_components=2)
self.transformed_points = nmf.fit_transform(self.points)
if self.version == "HDE-TruncatedSVD":
tsvd = TruncatedSVD(n_components=2)
self.transformed_points = tsvd.fit_transform(self.points)
if self.version == "HDE-LDA":
lda = LinearDiscriminantAnalysis(n_components=2)
y = []
for i in range(self.node_count):
y.append(1)
y = np.array(y)
lda = lda.fit(self.points, y=y)
self.transformed_points = lda.transform(self.points)
if self.version == "HDE-FR":
pca = PCA(n_components=2, copy=True)
self.transformed_points = pca.fit_transform(self.points)
if self.node_count == (self.edge_count +
1): # determine wether it is a tree.
FR = FR_Algorithm(number_of_nodes=self.node_count,
initial_temperature=self.initial_temperature,
cooling_factor=self.cooling_factor,
factor_attract=self.factor_attract,
factor_repulsion=self.factor_repulsion)
# use FR to fine-tune
self.transformed_points = FR.apply_force_directed_algorithm(
iteration=self.fr_iteration,
graph=self.graph,
coord_decomposition=self.transformed_points)
if self.version == "HDE-FICA-FR":
fica = FastICA(n_components=2)
self.transformed_points = fica.fit_transform(self.points)
if self.node_count == (self.edge_count +
1): # determine wether it is a tree.
FR = FR_Algorithm(number_of_nodes=self.node_count,
initial_temperature=self.initial_temperature,
cooling_factor=self.cooling_factor,
factor_attract=self.factor_attract,
factor_repulsion=self.factor_repulsion)
# use FR to fine-tune
self.transformed_points = FR.apply_force_directed_algorithm(
iteration=self.fr_iteration,
graph=self.graph,
coord_decomposition=self.transformed_points)
if self.version == "HDE-TSNE-FR":
# pca = PCA(n_components=10, copy=True)
# self.transformed_points = pca.fit_transform(self.points)
tsne = TSNE(learning_rate=self.learning_rate, init=self.init
) # 'init' must be 'pca', 'random', or a numpy array
self.transformed_points = tsne.fit_transform(self.points)
if self.node_count == (self.edge_count +
1): # determine wether it is a tree.
FR = FR_Algorithm(number_of_nodes=self.node_count,
initial_temperature=self.initial_temperature,
cooling_factor=self.cooling_factor,
factor_attract=self.factor_attract,
factor_repulsion=self.factor_repulsion)
# use FR to fine-tune
self.transformed_points = FR.apply_force_directed_algorithm(
iteration=self.fr_iteration,
graph=self.graph,
coord_decomposition=self.transformed_points)
if self.version == "HDE-SPE":
IP = SpectralEmbedding(n_components=2)
self.transformed_points = IP.fit_transform(self.distance_matrix)
# pca = PCA(n_components=2, copy=True)
# self.transformed_points = pca.fit_transform( self.transformed_points)
return self.node_count, self.edge_count
class SourceMethod():
def __init__(self, name, itrN, rg=[1, 8], err=0.1, method='pca'):
import xlrd
#print(name)
wb = xlrd.open_workbook(name)
st = wb.sheet_by_index(0)
self.xlsdata = []
for itr in range(st.nrows):
self.xlsdata.append((st.row_values(itr)))
self.rdata = []
self.sample = []
for itr in self.xlsdata:
self.rdata.append(itr[rg[0]:])
self.sample.append(itr[0])
self.title = self.xlsdata[0]
self.data_orig = np.abs(self.rdata[1:])
self.orig_data = np.array(self.rdata[1:])
self.dest_err = err
self.data_max = np.max(np.abs(self.data_orig), axis=0)
self.data = np.divide(self.data_orig, self.data_max)
self.minus = np.divide(self.orig_data[1, :],
np.abs(self.data_orig[1, :]))
self.itrN = itrN
self.train()
def func():
None
def train(self):
#data_avg=np.average(self.data)
data_min = np.min(np.abs(self.data), axis=0)
base = 0.5
data_sub = np.subtract(self.data, base * data_min)
#print(np.shape(data_min))
for itr in range(1, len(self.data)):
itr = 3
self.components_, self.array = mynmf(data_sub, itr, self.itrN,
self.dest_err)
self.method = FastICA(n_components=itr)
self.array = np.transpose(
self.method.fit_transform(np.transpose(data_sub)))
self.components_ = self.method.mixing_
self.array = np.add(self.array, data_min * base / itr)
#err=np.mean(np.abs(self.data-np.dot(self.components_,self.array)))
self.for_sta = np.multiply(
np.abs(np.dot(self.components_, self.array)), self.data_max)
sum_cof = 0
for itra in range(len(self.data[0])):
sum_cof += self.pearson(
np.transpose(self.data)[itra],
np.transpose(self.for_sta)[itra])
sum_cof = sum_cof / len(self.data[0])
print(sum_cof)
if (self.dest_err > 1 - sum_cof):
break
if (itr > 10):
break
#self.for_sta=np.multiply(np.abs(np.dot(self.components_,self.array)),self.data_max)
self.array = np.multiply(self.array, self.data_max)
self.data_orig = np.multiply(self.data_orig, self.minus)
self.for_sta = np.multiply(self.for_sta, self.minus)
self.array = np.multiply(self.array, self.minus)
self.data_orig = np.transpose(self.data_orig)
self.for_sta = np.transpose(self.for_sta)
self.array = np.transpose(self.array)
self.data = np.transpose(self.data)
self.orig_data = np.transpose(self.orig_data)
def get_par(self):
return self.array, self.method.n_components_, self.method.components_
def print_sta(self):
print("Source Number:")
print(self.components_)
ratio_sum = np.transpose([np.sum(self.method.components_, axis=1)])
print("Mixing ratio matrix:")
print(np.divide(self.method.components_, ratio_sum))
def pearson(self, x, y):
x_avg = np.average(x)
y_avg = np.average(y)
xv = x - x_avg
yv = y - y_avg
cof1 = np.sum(xv * yv)
x2 = np.sum(np.square(xv))
y2 = np.sum(np.square(yv))
if (y2 == 0):
cof = 0
else:
cof = cof1 / np.sqrt(x2 * y2)
return cof
def plot_sta(self):
import matplotlib.pyplot as plt
#plt.style.use('bmh')
plt.figure(1)
tick = np.arange(len(self.array))
width = 0.6 / len(self.array[0])
cont = 0
for dt in np.transpose(self.array):
#plt.plot(tick+width*cont,dt,alpha=0.4,color=list(plt.rcParams['axes.prop_cycle'])[cont]['color'])
#plt.bar(tick+width*cont,dt,width,alpha=0.2,color=list(plt.rcParams['axes.prop_cycle'])[cont]['color'])
cont = cont + 1
plt.figure(2)
cont = 0
for dt in np.transpose(self.array):
#plt.plot(dt,alpha=0.4,color=list(plt.rcParams['axes.prop_cycle'])[cont]['color'])
cont = cont + 1
plt.figure(3)
plt.plot(np.average(self.for_sta, axis=0))
for itr in range(len(self.data)):
plt.figure(4 + itr)
z1 = np.polyfit(self.data[itr], self.for_sta[itr], 1)
plt.scatter(self.data[itr], self.for_sta[itr])
mi = min(self.data[itr])
ma = max(self.data[itr])
idtv = ma - mi
x = np.arange(0, ma, 0.01)
plt.title("$" + str(self.title[itr]) + "$")
cof = self.pearson(self.data[itr], self.for_sta[itr])
#plt.text(mi/6+ma/6,idtv*0.001,"$f(x)=%fx+%f;cof=%f$"%(z1[0],z1[1],cof))
#plt.text(mi/1.9+ma/1.9,idtv*0.05,)
plt.plot(x, x * z1[0] + z1[1])
plt.boxplot(np.transpose(self.data_orig - self.for_sta))
plt.show()
emis = f["emis"][...] X = f["X"][...] f.close() TOL = 1e-4 emis[emis < TOL] = TOL emis[emis > 1 - TOL] = 1 - TOL ix = np.argsort(X) X = X[ix] emis = emis[ix, :] OD = -np.log(1 - emis) pcaOD = PCA(whiten=True, n_components=48) ica = FastICA(n_components=36, max_iter=5000) ODIR = ica.fit_transform(OD) # Reconstruct signals OD2 = ica.inverse_transform(ODIR) emis2 = 1 - np.exp(-OD2) # Reconstruct signals A_ = ica.mixing_ # Get estimated mixing matrix nmf = NMF(n_components=48) ODNR = nmf.fit_transform(OD) OD2 = nmf.inverse_transform(ODNR) emis2 = 1 - np.exp(-OD2) N = 48 knots = np.linspace(X.min(), X.max(), N)[1:-1] tck = splrep(X, -np.log(emis[:, 350]), t=knots) t = tck[0] c = np.zeros((emis.shape[-1], tck[1].size))
import numpy as np
import matplotlib.pyplot as plt
from scipy import signal
from sklearn.decomposition import PCA, FastICA
# Load data
input_file = 'mixture_of_signals.txt'
X = np.loadtxt(input_file)
# Compute ICA
ica = FastICA(n_components=4)
# Reconstruct the signals
signals_ica = ica.fit_transform(X)
# Get estimated mixing matrix
mixing_mat = ica.mixing_
# Perform PCA
pca = PCA(n_components=4)
signals_pca = pca.fit_transform(
X) # Reconstruct signals based on orthogonal components
# Specify parameters for output plots
models = [X, signals_ica, signals_pca]
colors = ['blue', 'red', 'black', 'green']
# Plotting input signal
plt.figure()
plt.title('Input signal (mixture)')
func = func_new
conv = signal.fftconvolve(func, emo, mode='same', axes=0)
conv /= np.linalg.norm(conv, axis=0)
conv = conv * 8
# PCA
X = np.concatenate((func, emo), axis=1)
pca = PCA(n_components=3, svd_solver='randomized')
pca.fit(X.transpose())
pca = pca.components_.transpose()
pca = pca * 10
# ICA
transformer = FastICA(n_components=3, random_state=0)
ica = transformer.fit_transform(X)
ica = ica * 8
# Fa
transformer = FactorAnalysis(n_components=3, random_state=0)
fa = transformer.fit_transform(X)
# GMM
from sklearn import mixture
gmmodel = mixture.GaussianMixture(n_components=3,
covariance_type='tied',
max_iter=100,
random_state=10).fit(X.transpose())
gmm = gmmodel.means_.transpose()
gmm_samp, gmm_y = gmmodel.sample(118)
def perform_feature_engineering(train, test, config):
for c in train.columns:
if (len(train[c].value_counts()) == 2):
if (train[c].mean() < config['SparseThreshold']):
del train[c]
del test[c]
col = list(test.columns)
if config['ID'] != True:
col.remove('ID')
# tSVD
if (config['tSVD'] == True):
tsvd = TruncatedSVD(n_components=config['n_comp'])
tsvd_results_train = tsvd.fit_transform(train[col])
tsvd_results_test = tsvd.transform(test[col])
for i in range(1, config['n_comp'] + 1):
train['tsvd_' + str(i)] = tsvd_results_train[:, i - 1]
test['tsvd_' + str(i)] = tsvd_results_test[:, i - 1]
# PCA
if (config['PCA'] == True):
pca = PCA(n_components=config['n_comp'])
pca2_results_train = pca.fit_transform(train[col])
pca2_results_test = pca.transform(test[col])
for i in range(1, config['n_comp'] + 1):
train['pca_' + str(i)] = pca2_results_train[:, i - 1]
test['pca_' + str(i)] = pca2_results_test[:, i - 1]
# ICA
if (config['ICA'] == True):
ica = FastICA(n_components=config['n_comp'])
ica2_results_train = ica.fit_transform(train[col])
ica2_results_test = ica.transform(test[col])
for i in range(1, config['n_comp'] + 1):
train['ica_' + str(i)] = ica2_results_train[:, i - 1]
test['ica_' + str(i)] = ica2_results_test[:, i - 1]
# GRP
if (config['GRP'] == True):
grp = GaussianRandomProjection(n_components=config['n_comp'], eps=0.1)
grp_results_train = grp.fit_transform(train[col])
grp_results_test = grp.transform(test[col])
for i in range(1, config['n_comp'] + 1):
train['grp_' + str(i)] = grp_results_train[:, i - 1]
test['grp_' + str(i)] = grp_results_test[:, i - 1]
# SRP
if (config['SRP'] == True):
srp = SparseRandomProjection(n_components=config['n_comp'],
dense_output=True,
random_state=420)
srp_results_train = srp.fit_transform(train[col])
srp_results_test = srp.transform(test[col])
for i in range(1, config['n_comp'] + 1):
train['srp_' + str(i)] = srp_results_train[:, i - 1]
test['srp_' + str(i)] = srp_results_test[:, i - 1]
if config['magic'] == True:
magic_mat = train[['ID', 'X0', 'y']]
magic_mat = magic_mat.groupby(['X0'])['y'].mean()
magic_mat = pd.DataFrame({
'X0': magic_mat.index,
'magic': list(magic_mat)
})
mean_magic = magic_mat['magic'].mean()
train = train.merge(magic_mat, on='X0', how='left')
test = test.merge(magic_mat, on='X0', how='left')
test['magic'] = test['magic'].fillna(mean_magic)
return train, test
sc = StandardScaler()
ica2 = FastICA(n_components = 2)
ica80 = FastICA(n_components = 80)
#Dataset 1
data1 = pd.read_csv('dist1.txt', sep = ' ')
data1.head()
data1 = data1.dropna(axis = 'index')
dataset1 = data1.values
data1_std = sc.fit_transform(dataset1)
data1_std_ica2 = ica2.fit_transform(data1_std)
data1_std_ica80 = ica80.fit_transform(data1_std)
data1_std_ica2
data1_std_ica80
plt.scatter(data1_std_ica2[:,0], data1_std_ica2[:,1])
dataset1frame80 = pd.DataFrame(data1_std_ica80)
dataset1frame80.head()
dataset1frame80['mean'] = dataset1frame80.mean(axis=1)
dataset1frame80.head()
ica_power1 = dataset1frame80['mean'].values
X = np.array(trainFile.data) Y = np.array(trainFile.labels) # just like the face recognition, we compute the avg digit image avg_digit = compute_avg_digits(X, configs.IMAGE_WIDTH) print "Avg digit computed ..." # Substract each input with the avg X_normalized_avg = normalize_with_avg(X, avg_digit) X_normalized = preprocessing.normalize(X_normalized_avg) print "Normalize X ..." # ICA Face ica = FastICA() features = ica.fit_transform(X_normalized) print "Transform done ..." # split into training and testing cutoff = len(Y) * 0.75 features_train = np.array(features[:cutoff]) Y_train = np.array(Y[:cutoff]) features_test = np.array(features[cutoff:]) Y_test = np.array(Y[cutoff:]) #Submission #features_train = np.array(features) #Y_train = np.array(Y) #X_test = np.array(testFile.data) #X_test_normalized_avg = normalize_with_avg(X_test, avg_digit) #X_test_normalized = preprocessing.normalize(X_test_normalized_avg)
def get_dc_feature(df_train,
df_test,
n_comp=12,
id_column=None,
label_column=None):
"""
构造分解特征
"""
train = df_train.copy()
test = df_test.copy()
if id_column:
train_id = train[id_column]
test_id = test[id_column]
train = drop_columns(train, [id_column])
test = drop_columns(test, [id_column])
if label_column:
train_y = train[label_column]
train = drop_columns(train, [label_column])
# tSVD
tsvd = TruncatedSVD(n_components=n_comp, random_state=420)
tsvd_results_train = tsvd.fit_transform(train)
tsvd_results_test = tsvd.transform(test)
# PCA
pca = PCA(n_components=n_comp, random_state=420)
pca2_results_train = pca.fit_transform(train)
pca2_results_test = pca.transform(test)
# ICA
ica = FastICA(n_components=n_comp, random_state=420)
ica2_results_train = ica.fit_transform(train)
ica2_results_test = ica.transform(test)
# GRP
grp = GaussianRandomProjection(n_components=n_comp,
eps=0.1,
random_state=420)
grp_results_train = grp.fit_transform(train)
grp_results_test = grp.transform(test)
# SRP
srp = SparseRandomProjection(n_components=n_comp,
dense_output=True,
random_state=420)
srp_results_train = srp.fit_transform(train)
srp_results_test = srp.transform(test)
# Append decomposition components to datasets
for i in range(1, n_comp + 1):
train['pca_' + str(i)] = pca2_results_train[:, i - 1]
test['pca_' + str(i)] = pca2_results_test[:, i - 1]
train['ica_' + str(i)] = ica2_results_train[:, i - 1]
test['ica_' + str(i)] = ica2_results_test[:, i - 1]
train['tsvd_' + str(i)] = tsvd_results_train[:, i - 1]
test['tsvd_' + str(i)] = tsvd_results_test[:, i - 1]
train['grp_' + str(i)] = grp_results_train[:, i - 1]
test['grp_' + str(i)] = grp_results_test[:, i - 1]
train['srp_' + str(i)] = srp_results_train[:, i - 1]
test['srp_' + str(i)] = srp_results_test[:, i - 1]
if id_column:
train[id_column] = train_id
test[id_column] = test_id
if label_column:
train[label_column] = train_y
return train, test
def update_EM(X, K, gamma, A, pi, mu, sigma_sqr, threshold=5e-5,
A_mode='GA', grad_mode='GA',
max_em_steps=30, n_gd_steps=20):
if type(X) is not torch.Tensor:
X = torch.tensor(X)
X = X.type(DTYPE).to(device)
N, D = X.shape
END = lambda dA, dsigma_sqr: (dA + dsigma_sqr) < threshold
Y = None
niters = 0
dA, dsigma_sqr = 10, 10
ret_time = {'E':[], 'obj':[]}
grad_norms, objs= [], []
if A_mode == 'random':
A = ortho_group.rvs(D)
A = to_tensor(A)
elif A_mode == 'ICA':
cov = X.T.matmul(X) / len(X)
cnt = 0
n_tries = 20
while cnt < n_tries:
try:
ica = FastICA()
_ = ica.fit_transform(X.cpu())
Aorig = ica.mixing_
# avoid numerical instability
U, ss, V = np.linalg.svd(Aorig)
ss /= ss[0]
ss[ss < SINGULAR_SMALL] = SINGULAR_SMALL
Aorig = (U * ss).dot(V)
A = np.linalg.inv(Aorig)
_, ss, _ = np.linalg.svd(A)
A = to_tensor(A / ss[0])
cnt = 2*n_tries
except:
cnt += 1
if cnt != 2*n_tries:
print('ICA failed. Use random.')
A = to_tensor(ortho_group.rvs(D))
while (not END(dA, dsigma_sqr)) and niters < max_em_steps:
niters += 1
A_prev, sigma_sqr_prev = A.clone(), sigma_sqr.clone()
objs += [],
if TIME: e_start = time()
Y, w, w_sumN, w_sumNK = E(X, A, pi, mu, sigma_sqr, Y=Y)
if TIME: ret_time['E'] += time() - e_start,
# M-step
if A_mode == 'ICA' or A_mode == 'None':
pi, mu, sigma_sqr = update_pi_mu_sigma(X, A, w, w_sumN, w_sumNK)
obj = get_objetive(X, A, pi, mu, sigma_sqr, w)
objs[-1] += obj,
if A_mode == 'CF': # gradient ascent
if CHECK_OBJ:
objs[-1] += get_objetive(X, A, pi, mu, sigma_sqr, w),
for i in range(n_gd_steps):
cf_start = time()
if VERBOSE: print(A.view(-1))
pi, mu, sigma_sqr = update_pi_mu_sigma(X, A, w, w_sumN, w_sumNK)
if TIME: a_start = time()
if grad_mode == 'CF1':
A = set_grad_zero(X, A, w, mu, sigma_sqr)
A = A.T
elif grad_mode == 'CF2':
cofs = get_cofactors(A)
det = torch.det(A)
if det < 0: # TODO: ignore neg det for now
cofs = cofs * -1
newA = A.clone()
for i in range(D):
for j in range(D):
t1 = (w[:, i] * X[:,j,None]**2 / sigma_sqr[i]).sum() / N
diff = (Y[i] - A[i,j] * X[:, j])[:, None] - mu[i]
t2 = (w[:, i] * X[:,j,None] * diff / sigma_sqr[i]).sum() / N
c1 = t1 * cofs[i,j]
c2 = t1 * (det - A[i,j]*cofs[i,j]) + t2 * cofs[i,j]
c3 = t2 * (det - A[i,j]*cofs[i,j]) - cofs[i,j]
inner = c2**2 - 4*c1*c3
if inner < 0:
print('Problme at solving for A[{},{}]: no real sol.'.format(i,j))
pdb.set_trace()
if c1 == 0:
sol = - c3 / c2
else:
sol = (inner**0.5 - c2) / (2*c1)
if False:
# check whether obj gets improved with each updated entry of A
curr_A = newA.clone()
curr_A[i,j] = sol
curr_obj = get_objetive(X, curr_A, pi, mu, sigma_sqr, w)
newA[i,j] = sol
A = newA.double()
# avoid numerical instability
U, ss, V = torch.svd(A)
ss = ss / ss[0]
ss[ss < SINGULAR_SMALL] = SINGULAR_SMALL
A = (U * ss).matmul(V)
if TIME:
if 'A' not in ret_time: ret_time['A'] = []
ret_time['A'] += time() - a_start,
if 'CF' not in ret_time: ret_time['CF'] = []
ret_time['CF'] += time() - cf_start,
if CHECK_OBJ:
if TIME: obj_start = time()
obj = get_objetive(X, A, pi, mu, sigma_sqr, w)
if TIME: ret_time['obj'] += time() - obj_start,
objs[-1] += obj,
if VERBOSE:
print('iter {}: obj= {:.5f}'.format(i, obj))
# pdb.set_trace()
# pdb.set_trace()
if A_mode == 'GA': # gradient ascent
if CHECK_OBJ:
objs[-1] += get_objetive(X, A, pi, mu, sigma_sqr, w),
for i in range(n_gd_steps):
ga_start = time()
if VERBOSE: print(A.view(-1))
pi, mu, sigma_sqr = update_pi_mu_sigma(X, A, w, w_sumN, w_sumNK)
if TIME: a_start = time()
# gradient steps
grad, y_time = get_grad(X, A, w, mu, sigma_sqr)
if TIME:
if 'Y' not in ret_time:
ret_time['Y'] = []
ret_time['Y'] += y_time,
if grad_mode == 'BTLS':
# backtracking line search
if TIME: obj_start = time()
obj = get_objetive(X, A, pi, mu, sigma_sqr, w)
if TIME: ret_time['obj'] += time() - obj_start,
beta, t, flag = 0.6, 1, True
gnorm = torch.norm(grad)
n_iter, ITER_LIM = 0, 10
while flag and n_iter < ITER_LIM:
n_iter += 1
Ap = A + t * grad
_, ss, _ = torch.svd(Ap)
Ap /= ss[0]
if TIME: obj_start = time()
obj_p = get_objetive(X, Ap, pi, mu, sigma_sqr, w)
if TIME: ret_time['obj'] += time() - obj_start,
t *= beta
base = obj - 0.5 * t * gnorm
flag = obj_p < base
gamma = t
ret_time['btls_nIters'] += n_iter,
elif grad_mode == 'perturb':
# perturb
perturb = A.std() * 0.1 * torch.randn(A.shape).type(DTYPE).to(device)
perturbed = A + perturb
perturbed_grad, _ = get_grad(X, perturbed, w, mu, sigma_sqr)
grad_diff = torch.norm(grad - perturbed_grad)
gamma = 1 /(EPS_GRAD + grad_diff) * 0.03
grad_norms += torch.norm(grad).item(),
A += gamma * grad
_, ss, _ = torch.svd(A)
A /= ss[0]
if TIME:
if 'A' not in ret_time: ret_time['A'] = []
ret_time['A'] += time() - a_start,
if 'GA' not in ret_time: ret_time['GA'] = []
ret_time['GA'] += time() - ga_start,
if CHECK_OBJ:
if TIME: obj_start = time()
obj = get_objetive(X, A, pi, mu, sigma_sqr, w)
if TIME: ret_time['obj'] += time() - obj_start,
objs[-1] += obj,
if VERBOSE:
print('iter {}: obj= {:.5f}'.format(i, obj))
# pdb.set_trace()
# pdb.set_trace()
if VERBOSE:
print('#{}: dA={:.3e} / dsigma_sqr={:.3e}'.format(niters, dA, dsigma_sqr))
print('A:', A.view(-1))
if TIME:
for key in ret_time:
ret_time[key] = np.array(ret_time[key]) if ret_time[key] else 0
# pdb.set_trace()
return A, pi, mu, sigma_sqr, grad_norms, objs, ret_time
# thus we need to use mask_strategy='epi' to compute the mask from the
# EPI images
masker = NiftiMasker(smoothing_fwhm=8,
memory='nilearn_cache',
memory_level=1,
mask_strategy='epi',
standardize=True)
data_masked = masker.fit_transform(func_filename)
#####################################################################
# Apply ICA
from sklearn.decomposition import FastICA
n_components = 10
ica = FastICA(n_components=n_components, random_state=42)
components_masked = ica.fit_transform(data_masked.T).T
# Normalize estimated components, for thresholding to make sense
components_masked -= components_masked.mean(axis=0)
components_masked /= components_masked.std(axis=0)
# Threshold
import numpy as np
components_masked[np.abs(components_masked) < .8] = 0
# Now invert the masking operation, going back to a full 3D
# representation
component_img = masker.inverse_transform(components_masked)
#####################################################################
# Visualize the results
def test_fastica_simple(add_noise, seed):
# Test the FastICA algorithm on very simple data.
rng = np.random.RandomState(seed)
# scipy.stats uses the global RNG:
n_samples = 1000
# Generate two sources:
s1 = (2 * np.sin(np.linspace(0, 100, n_samples)) > 0) - 1
s2 = stats.t.rvs(1, size=n_samples)
s = np.c_[s1, s2].T
center_and_norm(s)
s1, s2 = s
# Mixing angle
phi = 0.6
mixing = np.array([[np.cos(phi), np.sin(phi)], [np.sin(phi),
-np.cos(phi)]])
m = np.dot(mixing, s)
if add_noise:
m += 0.1 * rng.randn(2, 1000)
center_and_norm(m)
# function as fun arg
def g_test(x):
return x**3, (3 * x**2).mean(axis=-1)
algos = ["parallel", "deflation"]
nls = ["logcosh", "exp", "cube", g_test]
whitening = [True, False]
for algo, nl, whiten in itertools.product(algos, nls, whitening):
if whiten:
k_, mixing_, s_ = fastica(m.T,
fun=nl,
algorithm=algo,
random_state=rng)
with pytest.raises(ValueError):
fastica(m.T, fun=np.tanh, algorithm=algo)
else:
pca = PCA(n_components=2, whiten=True, random_state=rng)
X = pca.fit_transform(m.T)
k_, mixing_, s_ = fastica(X,
fun=nl,
algorithm=algo,
whiten=False,
random_state=rng)
with pytest.raises(ValueError):
fastica(X, fun=np.tanh, algorithm=algo)
s_ = s_.T
# Check that the mixing model described in the docstring holds:
if whiten:
assert_almost_equal(s_, np.dot(np.dot(mixing_, k_), m))
center_and_norm(s_)
s1_, s2_ = s_
# Check to see if the sources have been estimated
# in the wrong order
if abs(np.dot(s1_, s2)) > abs(np.dot(s1_, s1)):
s2_, s1_ = s_
s1_ *= np.sign(np.dot(s1_, s1))
s2_ *= np.sign(np.dot(s2_, s2))
# Check that we have estimated the original sources
if not add_noise:
assert_almost_equal(np.dot(s1_, s1) / n_samples, 1, decimal=2)
assert_almost_equal(np.dot(s2_, s2) / n_samples, 1, decimal=2)
else:
assert_almost_equal(np.dot(s1_, s1) / n_samples, 1, decimal=1)
assert_almost_equal(np.dot(s2_, s2) / n_samples, 1, decimal=1)
# Test FastICA class
_, _, sources_fun = fastica(m.T, fun=nl, algorithm=algo, random_state=seed)
ica = FastICA(fun=nl, algorithm=algo, random_state=seed)
sources = ica.fit_transform(m.T)
assert ica.components_.shape == (2, 2)
assert sources.shape == (1000, 2)
assert_array_almost_equal(sources_fun, sources)
assert_array_almost_equal(sources, ica.transform(m.T))
assert ica.mixing_.shape == (2, 2)
for fn in [np.tanh, "exp(-.5(x^2))"]:
ica = FastICA(fun=fn, algorithm=algo)
with pytest.raises(ValueError):
ica.fit(m.T)
with pytest.raises(TypeError):
FastICA(fun=range(10)).fit(m.T)
print('//===========================pca==========================')
pca = PCA(n)
traindata_pca = pca.fit_transform(traindata)
testdata_pca = pca.transform(testdata)
Faceidentifier(traindata_pca,trainlabel,testdata_pca,testlabel)
print('//===========================sfa==========================')
sfa = sfa.SFA()
traindata_sfa = sfa.fit_transform(traindata.T,conponents =n).T
testdata_sfa = sfa.transform(testdata.T).T
Faceidentifier(traindata_sfa,trainlabel,testdata_sfa,testlabel)
print('//===========================fastica==========================')
fastica = FastICA(n)
traindata_fastica = fastica.fit_transform(traindata)
testdata_fastica = fastica.transform(testdata)
Faceidentifier(traindata_fastica,trainlabel,testdata_fastica,testlabel)
for i in range(0,9):
if i == 0:
b = 0.1
elif i == 1:
b = 0.2
elif i == 2:
b = 0.5
elif i == 3:
b = 0.8
elif i == 4:
b = 1
elif i == 5:
