def joint_distribution_model(H):
numcontexts = H.shape[1]
counts = H.ravel()
size = counts.size
f0 = f_ssd(H.shape[0], H.shape[1])
F = np.asarray([f0])
return maxent.conditionalmodel(F, counts, numcontexts)
# Ideally, this could be stored as a sparse matrix of size C x X, whose ith row
# vector contains all points x_j in the sample space X in context c_i:
# N = sparse.lil_matrix((len(contexts), len(samplespace))) # initialized to zero
# for (c, x) in corpus:
# N[c, x] += 1
# This would be a nicer input format, but computations are more efficient
# internally with one long row vector. What we really need is for sparse
# matrices to offer a .reshape method so this conversion could be done
# internally and transparently. Then the numcontexts argument to the
# conditionalmodel constructor could also be inferred from the matrix
# dimensions.
# Create a model
model = maxentropy.conditionalmodel(F, N, numcontexts)
model.verbose = True
# Fit the model
model.fit()
# Output the distribution
print "\nFitted model parameters are:\n" + str(model.params)
p = model.probdist()
print "\npmf table p(x | c), where c is the context 'the':"
c = contexts.index('the')
print p[c * numsamplepoints:(c + 1) * numsamplepoints]
# Ideally, this could be stored as a sparse matrix of size C x X, whose ith row
# vector contains all points x_j in the sample space X in context c_i:
# N = sparse.lil_matrix((len(contexts), len(samplespace))) # initialized to zero
# for (c, x) in corpus:
# N[c, x] += 1
# This would be a nicer input format, but computations are more efficient
# internally with one long row vector. What we really need is for sparse
# matrices to offer a .reshape method so this conversion could be done
# internally and transparently. Then the numcontexts argument to the
# conditionalmodel constructor could also be inferred from the matrix
# dimensions.
# Create a model
model = maxentropy.conditionalmodel(F, N, numcontexts)
model.verbose = True
# Fit the model
model.fit()
# Output the distribution
print "\nFitted model parameters are:\n" + str(model.params)
p = model.probdist()
print "\npmf table p(x | c), where c is the context 'the':"
c = contexts.index("the")
print p[c * numsamplepoints : (c + 1) * numsamplepoints]
