def __init__(self, data, index=None, pca=False, min_std=10e-9):
self.data = data
self._pca = None
self._zs = ZScore(data, replace_inf=True)
data = self._zs.normalize(data)
# to remove columns that contaim nan's and have zero variance
mask_nan = np.invert(np.isnan(data.sum(axis=0)))
self._mask = np.ones(mask_nan.shape).astype(bool)
if index is not None:
self._mask[:] = False
self._mask[index] = True
self._mask *= mask_nan
data = self.filter(data)
if pca:
# data1 = data self.normalize(data)
self._pca = PCA(data, minfrac=0.05)
self.traindata = self._pca.project(data)
else:
self.traindata = data #self.normalize(data)
def __call__(self):
# z-scoring
if self.treedata is None:
raise SortingError("No examples for similarity measure available!")
zs = ZScore(self.data, replace_inf=True)
data_zs = zs.normalize(self.data)
# z-scored mean value of pca procjected treedata
mu = zs.normalize(self.treedata)
data_zs, mu = filter_nans(data_zs, mu)
mu = mu.mean(axis=0)
distsq = [np.power((x - mu), 2).sum() for x in data_zs]
return np.sqrt(distsq)
def __call__(self):
# z-scoring
if self.treedata is None:
raise SortingError("No examples for similarity measure available!")
if self.treedata.shape[1] < 2:
raise SortingError(("CosineSimilarity needs at least 2 "
"features for sorting"))
zs = ZScore(self.data, replace_inf=True)
data_zs = zs.normalize(self.data)
# z-scored mean value of pca procjected treedata
mu = zs.normalize(self.treedata)
data_zs, mu = filter_nans(data_zs, mu)
mu = mu.mean(axis=0)
denom = np.sqrt((mu**2).sum()) * np.sqrt((data_zs**2).sum(axis=1))
s_cos = np.sum(mu * data_zs, axis=1) / denom
return -1.0 * s_cos
