block structure in transition matrices

# A = model.trans_distn.A.copy()

labels_norep, _ = pymouse.get_norep(np.hstack(model.stateseqs))
A = build_frequentist_statemap(labels_norep)

import spectral_clustering.spclust as sp
labels_norep, _ = pymouse.get_norep(np.hstack(model.stateseqs))
used_labels = np.argwhere(np.bincount(labels_norep).astype('float32')/len(labels_norep) > 0.001).ravel()
A = A[np.ix_(used_labels,used_labels)]

d = np.repeat(A.sum(1)[:,np.newaxis], len(A), 1)
d[d==0] = 1.0
A = A/d

n_blocks = 7
n_eigenvectors = 3

blockNode = sp.SpectralBlockify(n_blocks=n_blocks, n_eigenvectors=n_eigenvectors)
blockNode.fit(mean_first_passage(A))
perm_A = blockNode.permute(A)

import pymouse.entropy
print pymouse.entropy.statemap_entropy(A)
figure(figsize=(5,5))
imshow(perm_A, interpolation='nearest')