def getAsymmetricSigmaForScipyMatrix(raw_mean_control_profile, dev_control_matrix, config_params):
dev_control_matrix_tall = np.zeros((0, 1))
repeated_raw_mean_control_profile = np.zeros((0, 1))
raw_mean_control_profile = np.matrix(raw_mean_control_profile).transpose()
for i in range(dev_control_matrix.shape[1]):
dev_control_matrix_tall = np.vstack((dev_control_matrix_tall, dev_control_matrix[:, [i]]))
repeated_raw_mean_control_profile = np.vstack((repeated_raw_mean_control_profile, raw_mean_control_profile))
dev_control_matrix_tall = np.array(dev_control_matrix_tall)
pos = dev_control_matrix_tall >= 0
neg = dev_control_matrix_tall < 0
dev_control_matrix_tall_squared = dev_control_matrix_tall**2
lowess = np.zeros(dev_control_matrix_tall.shape)
for i in range(dev_control_matrix.shape[1]):
if get_verbosity(config_params) >= 3:
print dev_control_matrix_tall[pos], dev_control_matrix_tall_squared[pos]
print dev_control_matrix_tall.shape, dev_control_matrix_tall_squared.shape, pos.shape, scipy.nonzero(pos)[0].shape[0]
lowess[pos] = np.array( mlab.smooth(repeated_raw_mean_control_profile[pos], dev_control_matrix_tall_squared[pos] , 0.3, 'lowess') ).transpose()[0]
lowess[neg] = np.array( mlab.smooth(repeated_raw_mean_control_profile[neg], dev_control_matrix_tall_squared[neg] , 0.3, 'lowess') ).transpose()[0]
lowess_pos = np.zeros(raw_mean_control_profile.shape) + np.nan
lowess_neg = np.zeros(raw_mean_control_profile.shape) + np.nan
for i in range(raw_mean_control_profile.shape[0]):
for j in range(i, dev_control_matrix_tall.shape[0], raw_mean_control_profile.shape[0]):
if pos[j]:
lowess_pos[i] = lowess[j]
else:
lowess_neg[i] = lowess[j]
lowess_symmetric = np.array( mlab.smooth(repeated_raw_mean_control_profile, dev_control_matrix_tall, 'lowess') ).transpose()[0]
return np.sqrt( lowess_neg ).real , np.sqrt( lowess_pos ).real, np.sqrt(lowess_symmetric[range(raw_mean_control_profile.shape[0])]).real
def smooth(xx, yy):
k = wellbehaved(xx) & wellbehaved(yy)
yy_normalized = np.zeros(xx.shape) + np.nan
if np.sum(yy[k]) == 0:
return yy_normalized
lowess_line = np.array( mlab.smooth(xx[k], yy[k], 0.3, 'rlowess') ).transpose()[0]
if np.sum(lowess_line) == 0:
return yy_normalized
yy_normalized[k] = xx[k] * yy[k] / lowess_line
return yy_normalized
