def pk_questionnaire(matrix,iters):
col_data = matrix
col_aff = qmain.local_geometry_norm_ip(col_data)
p,eigvals,eigvecs = cdiff.diffusion_embed(col_aff,normalized=True)
n_eigs = np.sum(np.abs(eigvals[~np.isnan(eigvals)]) > 1e-14)
tree_cols = cdiff.cluster(eigvecs[:,1:n_eigs],eigvals[1:n_eigs])
#tree_cols.disp_tree()
row_data = qmain.extend_coords_means(matrix.T,tree_cols,False)
row_aff = qmain.local_geometry_norm_ip(row_data)
p,eigvals,eigvecs = cdiff.diffusion_embed(row_aff,normalized=True)
n_eigs = np.sum(np.abs(eigvals[~np.isnan(eigvals)]) > 1e-14)
tree_rows = cdiff.cluster(eigvecs[:,1:n_eigs],eigvals[1:n_eigs])
#tree_rows.disp_tree()
for i in xrange(iters):
col_data = qmain.extend_coords_means(matrix,tree_rows,False,False)
col_aff = qmain.local_geometry_gaussian(col_data,knn=100)
p,col_eigvals,col_eigvecs = cdiff.diffusion_embed(col_aff,normalized=True)
n_eigs = np.sum(np.abs(col_eigvals[~np.isnan(col_eigvals)]) > 1e-14)
tree_cols = cdiff.cluster(col_eigvecs[:,1:n_eigs],col_eigvals[1:n_eigs])
#tree_cols.disp_tree()
row_data = qmain.extend_coords_means(matrix.T,tree_cols,False,False)
row_aff = qmain.local_geometry_gaussian(row_data,knn=75)
p,row_eigvals,row_eigvecs = cdiff.diffusion_embed(row_aff,normalized=True)
n_eigs = np.sum(np.abs(row_eigvals[~np.isnan(row_eigvals)]) > 1e-14)
tree_rows = cdiff.cluster(row_eigvecs[:,1:n_eigs],row_eigvals[1:n_eigs])
#tree_rows.disp_tree()
return tree_rows,tree_cols,row_eigvecs,col_eigvecs
def mmpi_questionnaire(matrix,iters):
col_data = matrix
col_aff = qmain.local_geometry_gaussian(col_data,knn=COLS_KNN,
init_eps_pt=False,symm_inner=True)
p,eigvals,eigvecs = cdiff.diffusion_embed(col_aff,n_eigs=N_COL_EIGS,
normalized=True)
n_eigs = np.sum(np.abs(eigvals[~np.isnan(eigvals)]) > 1e-14)
tree_cols = cdiff.cluster(eigvecs[:,1:n_eigs],eigvals[1:n_eigs],
knn=CLUSTER_KNN)
#tree_cols.disp_tree()
row_data = qmain.extend_coords_means(matrix.T,tree_cols,
weighted=WEIGHTED,
singletons=SINGLETONS)
#row_aff = qmain.local_geometry_norm_ip(row_data)
row_aff = qmain.local_geometry_gaussian(row_data,knn=ROWS_KNN,
init_eps_pt=False,symm_inner=True)
p,eigvals,eigvecs = cdiff.diffusion_embed(row_aff,n_eigs=N_ROW_EIGS,
normalized=True)
n_eigs = np.sum(np.abs(eigvals[~np.isnan(eigvals)]) > 1e-14)
tree_rows = cdiff.cluster(eigvecs[:,1:n_eigs],eigvals[1:n_eigs],
knn=CLUSTER_KNN)
#tree_rows.disp_tree()
for i in xrange(iters):
col_data = qmain.extend_coords_means(matrix,tree_rows,
weighted=WEIGHTED,
singletons=SINGLETONS)
col_aff = qmain.local_geometry_gaussian(col_data,knn=COLS_KNN,
init_eps_pt=False,symm_inner=False)
p,col_eigvals,col_eigvecs = cdiff.diffusion_embed(col_aff,
n_eigs=N_COL_EIGS,
normalized=True)
#print col_eigvals
n_eigs = np.sum(np.abs(col_eigvals[~np.isnan(col_eigvals)]) > 1e-14)
tree_cols = cdiff.cluster(col_eigvecs[:,1:n_eigs],col_eigvals[1:n_eigs],
knn=CLUSTER_KNN)
#tree_cols.disp_tree()
row_data = qmain.extend_coords_means(matrix.T,tree_cols,WEIGHTED,SINGLETONS)
row_aff = qmain.local_geometry_gaussian(row_data,knn=ROWS_KNN,
init_eps_pt=False,symm_inner=False)
p,row_eigvals,row_eigvecs = cdiff.diffusion_embed(row_aff,
n_eigs=N_ROW_EIGS,
normalized=True)
#print row_eigvals
n_eigs = np.sum(np.abs(row_eigvals[~np.isnan(row_eigvals)]) > 1e-14)
tree_rows = cdiff.cluster(row_eigvecs[:,1:n_eigs],row_eigvals[1:n_eigs],
knn=CLUSTER_KNN)
#tree_rows.disp_tree()
return tree_rows,tree_cols,row_eigvecs,col_eigvecs,row_eigvals,col_eigvals
import qmain
import cluster_diffusion as cdiff
import numpy as np
import tree_utils
reload(qmain)
reload(cdiff)
reload(tree_utils)
matrix = np.zeros([128,256])
for i in xrange(128):
for j in xrange(256):
matrix[i,j] = np.sin(i*j*np.pi/256.0)
affinity = qmain.local_geometry_gaussian(matrix, 10)
p,eigvals,eigvecs = cdiff.diffusion_embed(affinity,normalized=True)
tree_cols = cdiff.cluster(eigvecs[:,1:],eigvals[1:])
#dual_affinity = qmain.dual_geometry_means(matrix.T, tree_cols, 4)
#p,eigvals,eigvecs = cdiff.diffusion_embed(dual_affinity,normalized=True)
#tree_rows = cdiff.cluster(eigvecs[:,1:],eigvals[1:])
new_aff = qmain.dual_geometry_norm_ip_max(matrix.T, tree_cols)
p,eigvals,eigvecs = cdiff.diffusion_embed(new_aff,normalized=True)
tree_rows = cdiff.cluster(eigvecs[:,1:],eigvals[1:])
new_col_aff = qmain.dual_geometry_norm_ip_max(matrix, tree_rows)
p,eigvals,eigvecs = cdiff.diffusion_embed(new_col_aff,normalized=True)
tree_cols2 = cdiff.cluster(eigvecs[:,1:],eigvals[1:])
new_row_aff = qmain.dual_geometry_norm_ip_max(matrix.T, tree_cols2)
p,eigvals,eigvecs = cdiff.diffusion_embed(new_row_aff,normalized=True)