N = 40 # estimated number of neurons
TargetArea = N * prod(2 * array(sig)) / prod(data[0, :, :].shape)
TargetRange = [TargetArea * 0.8, TargetArea * 1.2]
NonNegative = True
lam = 1
elif data_source == 3: # Use experimental 3D data
mat = loadmat('Datasets/data_exp3D')
data = transpose(mat['data'], [3, 0, 1, 2])
sig = (2, 2, 2) # neurons size
TargetRange = [0.005, 0.015]
NonNegative = True
lam = 0.001
# Run source detection algorithms
x = gaussian_group_lasso(data, sig, lam,
NonNegative=NonNegative, TargetAreaRatio=TargetRange, verbose=True, adaptBias=True)
# x = gaussian_group_lasso(data[:len(data) / 5 * 5].reshape((-1, 5) + data.shape[1:]).max(1), sig, lam/5.,
# NonNegative=NonNegative, TargetAreaRatio=TargetRange, verbose=True, adaptBias=True)
pic_x = percentile(x, 95, axis=0)
pic_data = percentile(data, 95, axis=0)
# centers extracted from fista output using RegionalMax
cent = GetCenters(pic_x)
MSE_array, shapes, activity, boxes = LocalNMF(
data, (array(cent)[:-1]).T, sig,
NonNegative=NonNegative, verbose=True, adaptBias=True)
L = len(cent[0]) # number of detected neurons
denoised_data = activity[:L].T.dot(shapes[:L].reshape(L, -1)).reshape(data.shape)
pic_denoised = percentile(denoised_data, 95, axis=0)
residual = data - activity.T.dot(shapes.reshape(len(shapes), -1)).reshape(data.shape)
def GetCentersData(data,NumCent,data_name=[],rep=0):
"""
Get intialization centers using group lasso
Input
----------
data : array, shape (T, X,Y,(,Z))
data
data_name: string
dataset name so we can save load previous center data
NumCent: integer
number centers to extract
rep: integer
repetition number
Output
----------
activity: array, shape (L,T)
extracted temporal components
"""
from numpy import array,percentile
from BlockGroupLasso import gaussian_group_lasso, GetCenters
from pylab import load
import os
import pickle
DataFolder=GetDataFolder()
center_file_name=DataFolder + '/centers_'+ str(data_name) + '_rep_' + str(rep)
if NumCent>0:
if data_name==[] or os.path.isfile(center_file_name)==False:
if data.ndim==3:
sig0=(2,2)
else:
sig0=(2,2,2)
TargetRange = [0.1, 0.2]
lam = 500
ds= 50 #downscale time for group lasso
NonNegative=True
downscaled_data=data[:int(old_div(len(data), ds)) * ds].reshape((-1, ds) + data.shape[1:]).max(1) #for speed ups
x = gaussian_group_lasso(downscaled_data, sig0, lam,NonNegative=NonNegative, TargetAreaRatio=TargetRange, verbose=True, adaptBias=False)
pic_x = percentile(x, 95, axis=0)
######
# centers extracted from fista output using RegionalMax
cent = GetCenters(pic_x)
print(np.shape(cent)[0])
# Plot Results
# import matplotlib.pyplot as plt
# pic_data = np.percentile(data, 95, axis=0)
# plt.figure(figsize=(12, 4. * data.shape[1] / data.shape[2]))
# ax = plt.subplot(131)
# ax.scatter(cent[1], cent[0], marker='o', c='white')
# plt.hold(True)
# ax.set_title('Data + centers')
# ax.imshow(pic_data.max(2))
# ax2 = plt.subplot(132)
# ax2.scatter(cent[1], cent[0], marker='o', c='white')
# ax2.imshow(pic_x.max(2))
# ax2.set_title('Inferred x')
# ax3 = plt.subplot(133)
# ax3.scatter(cent[1], cent[0], marker='o', c='white')
# ax3.imshow(pic_x.max(2))
# ax3.set_title('Denoised data')
# plt.show()
#
# save results
if data_name!=[]:
from io import open
f = open(center_file_name, 'wb')
pickle.dump(cent, f, protocol=pickle.HIGHEST_PROTOCOL)
f.close()
else:
if data_name!=[]:
cent=load(center_file_name)
new_cent=(array(cent)[:-1]).T
new_cent=new_cent[:NumCent] #just give strongest centers
else:
new_cent=np.reshape([],(0,data.ndim-1))
return new_cent
def GetCentersData(data, data_name, NumCent):
from numpy import array, percentile
from BlockGroupLasso import gaussian_group_lasso, GetCenters
from pylab import load
import matplotlib.pyplot as plt
import os
import cPickle
DataFolder = GetDataFolder()
center_file_name = DataFolder + '/centers_' + data_name
if NumCent > 0:
if os.path.isfile(center_file_name) == False:
if data.ndim == 3:
sig0 = (2, 2)
else:
sig0 = (2, 2, 2)
TargetRange = [0.1, 0.2]
lam = 500
ds = 50 #downscale time for group lasso
NonNegative = True
downscaled_data = data[:int(len(data) / ds) *
ds].reshape((-1, ds) +
data.shape[1:]).max(1)
x = gaussian_group_lasso(downscaled_data,
sig0,
lam,
NonNegative=NonNegative,
TargetAreaRatio=TargetRange,
verbose=True,
adaptBias=False)
pic_x = percentile(x, 95, axis=0)
######
# centers extracted from fista output using RegionalMax
cent = GetCenters(pic_x)
print np.shape(cent)[0]
# Plot Results
# pic_data = np.percentile(data, 95, axis=0)
# plt.figure(figsize=(12, 4. * data.shape[1] / data.shape[2]))
# ax = plt.subplot(131)
# ax.scatter(cent[1], cent[0], marker='o', c='white')
# plt.hold(True)
# ax.set_title('Data + centers')
# ax.imshow(pic_data.max(2))
# ax2 = plt.subplot(132)
# ax2.scatter(cent[1], cent[0], marker='o', c='white')
# ax2.imshow(pic_x.max(2))
# ax2.set_title('Inferred x')
# ax3 = plt.subplot(133)
# ax3.scatter(cent[1], cent[0], marker='o', c='white')
# ax3.imshow(pic_x.max(2))
# ax3.set_title('Denoised data')
# plt.show()
#
f = file(center_file_name, 'wb')
cPickle.dump(cent, f, protocol=cPickle.HIGHEST_PROTOCOL)
f.close()
else:
cent = load(center_file_name)
new_cent = (array(cent)[:-1]).T
new_cent = new_cent[:NumCent] #just give strongest centers
else:
new_cent = np.reshape([], (0, data.ndim - 1))
return new_cent
N = 40 # estimated number of neurons
TargetArea = N * prod(2 * array(sig)) / prod(data[0, :, :].shape)
TargetRange = [TargetArea * 0.8, TargetArea * 1.2]
NonNegative = True
lam = 1
elif data_source == 3: # Use experimental 3D data
mat = loadmat('Datasets/data_exp3D')
data = transpose(mat['data'], [3, 0, 1, 2])
sig = (2, 2, 2) # neurons size
TargetRange = [0.005, 0.015]
NonNegative = True
lam = 0.001
# Run source detection algorithms
x = gaussian_group_lasso(data, sig, lam, NonNegative=NonNegative,
TargetAreaRatio=TargetRange, verbose=True, adaptBias=True)
pic_x = percentile(x, 95, axis=0)
pic_data = percentile(data, 95, axis=0)
# centers extracted from fista output using RegionalMax
cent = GetCenters(pic_x)
# ROI around each center, using watersheding on non-zero regions
ROI = GetROI(pic_x, (array(cent)[:-1]).T)
# temporal traces of activity for each neuron, averaged over each ROI
activity = GetActivity(x, ROI)
MSE_array, shapes, activity, boxes, background = LocalNMF(
data, (array(cent)[:-1]).T, activity, sig,
NonNegative=NonNegative, verbose=True, adaptBias=True)
L = len(shapes) # number of detected neurons
denoised_data = 0 * data
def GetCentersData(data,data_name,NumCent):
from numpy import array,percentile
from BlockGroupLasso import gaussian_group_lasso, GetCenters
from pylab import load
import matplotlib.pyplot as plt
import os
import cPickle
DataFolder=GetDataFolder()
center_file_name=DataFolder + '/centers_'+data_name
if NumCent>0:
if os.path.isfile(center_file_name)==False:
if data.ndim==3:
sig0=(2,2)
else:
sig0=(2,2,2)
TargetRange = [0.1, 0.2]
lam = 500
ds= 50 #downscale time for group lasso
NonNegative=True
downscaled_data=data[:int(len(data) / ds) * ds].reshape((-1, ds) + data.shape[1:]).max(1)
x = gaussian_group_lasso(downscaled_data, sig0, lam,NonNegative=NonNegative, TargetAreaRatio=TargetRange, verbose=True, adaptBias=False)
pic_x = percentile(x, 95, axis=0)
######
# centers extracted from fista output using RegionalMax
cent = GetCenters(pic_x)
print np.shape(cent)[0]
# Plot Results
# pic_data = np.percentile(data, 95, axis=0)
# plt.figure(figsize=(12, 4. * data.shape[1] / data.shape[2]))
# ax = plt.subplot(131)
# ax.scatter(cent[1], cent[0], marker='o', c='white')
# plt.hold(True)
# ax.set_title('Data + centers')
# ax.imshow(pic_data.max(2))
# ax2 = plt.subplot(132)
# ax2.scatter(cent[1], cent[0], marker='o', c='white')
# ax2.imshow(pic_x.max(2))
# ax2.set_title('Inferred x')
# ax3 = plt.subplot(133)
# ax3.scatter(cent[1], cent[0], marker='o', c='white')
# ax3.imshow(pic_x.max(2))
# ax3.set_title('Denoised data')
# plt.show()
#
f = file(center_file_name, 'wb')
cPickle.dump(cent, f, protocol=cPickle.HIGHEST_PROTOCOL)
f.close()
else:
cent=load(center_file_name)
new_cent=(array(cent)[:-1]).T
new_cent=new_cent[:NumCent] #just give strongest centers
else:
new_cent=np.reshape([],(0,data.ndim-1))
return new_cent
