def _aux(data, settings):
mu, var = temporal_moments_3D(data, settings, axis=0).T
pvals = _bivariate_student_orderability_from_moments(mu,
var,
len(data),
type="pval")
return gmean(offdiag_1D(pvals))
def _aux(data2DLst, settings):
# baselines = _get_baselines_non_uniform(data2DLst, settings)
mu, var = temporal_moments_3D_non_uniform(data2DLst, settings,
axis=0).T
pvals = _bivariate_student_orderability_from_moments(mu,
var,
len(data2DLst),
type="pval")
return gmean(offdiag_1D(pvals))
def _bivariate_binary_orderability_from_moments(mu):
'''
For each pair of channels, decide which one is earlier by comparing their time courses, converted to CDF
:param mu: 1D Array of temporal means for every channel
:return: 2D array of shape [nCell, nCell],
'''
biasRowExtrude = np.tile(mu, (len(mu), 1))
rez = biasRowExtrude - biasRowExtrude.T
# Check that there are no off-diagonal differences that evaluated exactly to zero.
assert np.sum(offdiag_1D(rez) == 0) == 0
return rez > 0
def cluster_values(M, clusters):
'''
:param M: Distance matrix
:param clusters: Cluster labels
:return: Dictionary of metrics of cluster fidelity. Each metric computed for each cluster
avgCorr - average in-cluster distance
avgRelCorr - average in-cluster distance minus out-cluster distance
'''
rez = {'avgCorr': {}, 'avgRelCorr': {}}
for iClust in sorted(set(clusters)):
idxClust = clusters == iClust
clustBlock = M[idxClust][:, idxClust]
clustNonclustBlock = M[~idxClust][:, idxClust]
muClust = np.mean(offdiag_1D(clustBlock))
muNon = np.mean(clustNonclustBlock)
rez['avgCorr'][iClust] = muClust
rez['avgRelCorr'][iClust] = muClust - muNon
return rez
def avg_bivariate_binary_orderability_from_temporal_mean(temporalMeans2D):
return np.nanmean(
offdiag_1D(
bivariate_orderability_from_temporal_mean(temporalMeans2D, {})))
def avg_bivariate_binary_orderability_3D_non_uniform(data2DLst, settings):
return np.nanmean(
offdiag_1D(
bivariate_binary_orderability_3D_non_uniform(data2DLst, settings)))
def avg_bivariate_binary_orderability_3D(data, settings):
return np.nanmean(
offdiag_1D(bivariate_binary_orderability_3D(data, settings)))
def avg_corr_3D_non_uniform(dataLst, settings):
M = corr_3D_non_uniform(dataLst, settings)
return np.nanmean(np.abs(offdiag_1D(M)))
def avg_corr_3D(data, settings):
M = corr_3D(data, settings)
return np.nanmean(np.abs(offdiag_1D(M)))