def findAnchors(Q, K, params, candidates=None):
eps = params['eps']
# Random number generator for generating dimension reduction
if params['seed'] > 0:
prng_W = RandomState(params['seed'])
else:
prng_W = RandomState(None)
#checkpoint_prefix = params['checkpoint_prefix']
new_dim = params['new_dim']
if candidates == None:
candidates = np.arange(Q.shape[0])
# row normalize Q
row_sums = Q.sum(1)
row_sums[row_sums < eps] = eps
for i in xrange(len(Q[:, 0])):
Q[i, :] = Q[i, :]/float(row_sums[i])
# Reduced dimension random projection method for recovering anchor words
Q_red = rp.Random_Projection(Q.T, new_dim, prng_W)
Q_red = Q_red.T
(anchors, anchor_indices) = gs.Projection_Find(Q_red, K, candidates)
# restore the original Q
for i in xrange(len(Q[:, 0])):
Q[i, :] = Q[i, :]*float(row_sums[i])
return anchor_indices
def find_anchors(Q, K, candidates, dim, seed):
# Random number generator for generating dimension reduction
prng_W = np.random.RandomState(seed)
# row normalize Q
row_sums = Q.sum(1)
for i in range(len(Q[:, 0])):
Q[i, :] = Q[i, :] / float(row_sums[i])
# Reduced dimension random projection method for recovering anchor words
Q_red = rp.Random_Projection(Q.T, dim, prng_W)
Q_red = Q_red.T
(anchors, anchor_indices) = gs.Projection_Find(Q_red, K, candidates)
# restore the original Q
for i in range(len(Q[:, 0])):
Q[i, :] = Q[i, :] * float(row_sums[i])
return anchor_indices
def findAnchors(Q, K, params, candidates):
# Random number generator for generating dimension reduction
prng_W = RandomState(params.seed)
checkpoint_prefix = params.checkpoint_prefix
new_dim = params.new_dim
# row normalize Q
row_sums = Q.sum(1)
for i in xrange(len(Q[:, 0])):
Q[i, :] = Q[i, :] / float(row_sums[i])
# Reduced dimension random projection method for recovering anchor words
Q_red = rp.Random_Projection(Q.T, new_dim, prng_W)
Q_red = Q_red.T
(anchors, anchor_indices) = gs.Projection_Find(Q_red, K, candidates)
# restore the original Q
for i in xrange(len(Q[:, 0])):
Q[i, :] = Q[i, :] * float(row_sums[i])
return anchor_indices
def findAnchors(Q, K, params, candidates):
# row normalize Q
row_sums = Q.sum(axis=1)
for i in xrange(len(Q[:, 0])):
Q[i, :] = Q[i, :]/float(row_sums[i] + 1e-100)
# Reduced dimension random projection method for recovering anchor words
if params.lowerDim is None or params.lowerDim >= Q.shape[1]:
Q_red = Q.copy()
else:
# Random number generator for generating dimension reduction
prng_W = RandomState(params.seed)
Q_red = rp.Random_Projection(Q.T, params.lowerDim, prng_W)
Q_red = Q_red.T
(anchors, anchor_indices) = gs.Projection_Find(Q_red, K, candidates)
# restore the original Q
for i in xrange(len(Q[:, 0])):
Q[i, :] = Q[i, :]*float(row_sums[i])
return anchor_indices
R = rp.Random_Matrix(V, params.new_dim, prng)
#only accept anchors that appear in a significant number of docs
print "identifying candidate anchors"
candidate_anchors = []
for i in xrange(V):
if len(np.nonzero(row_M[i, :])[1]) > params.anchor_thresh:
candidate_anchors.append(i)
print len(candidate_anchors), "candidates"
Q = np.vstack(
generate_Q_matrix(row_M,
col_M,
row_normalize=True,
projection_matrix=R.T)) #row-by-row generation
_, anchors = gs.Projection_Find(Q, K, candidate_anchors)
print "anchors are:", anchors
anchor_file = file(outfile + '.anchors', 'w')
print >> anchor_file, "\t".join(["topic id", "word id", "word"])
for i, a in enumerate(anchors):
print i, vocab[a]
print >> anchor_file, "\t".join([str(x) for x in (i, a, vocab[a])])
anchor_file.close()
#recover topics
row_sums = np.array(row_M.sum(1)).reshape(V)
#generate Q_matrix rows for anchors
Q_A = np.vstack(
generate_Q_matrix(row_M,
col_M,