def gen_partition_from_weights(n_rows, n_cols, view_weights, clusters_weights):
n_views = len(view_weights)
Zv = [v for v in range(n_views)]
for _ in range(n_cols-n_views):
v = utils.pflip(view_weights)
Zv.append(v)
random.shuffle(Zv)
assert len(Zv) == n_cols
Zc = []
for v in range(n_views):
n_clusters = len(clusters_weights[v])
Z = [c for c in range(n_clusters)]
for _ in range(n_rows-n_clusters):
c_weights = numpy.copy(clusters_weights[v])
c = utils.pflip(c_weights)
Z.append(c)
random.shuffle(Z)
Zc.append(Z)
assert len(Zc) == n_views
assert len(Zc[0]) == n_rows
return Zv, Zc
def _gen_multinomial_data_column(Z, separation=.9, distargs=None):
n_rows = len(Z)
K = distargs['K']
if separation > .95:
separation = .95
Tc = numpy.zeros(n_rows, dtype=int)
C = max(Z)+1
theta_arrays = [numpy.random.dirichlet(numpy.ones(K)*(1.0-separation), 1) for _ in range(C)]
for r in range(n_rows):
cluster = Z[r]
thetas = theta_arrays[cluster][0]
x = int(utils.pflip(thetas))
Tc[r] = x
return Tc
def _simple_predictive_sample_unobserved(state, cols, N=1):
# get if in the same view
views = [state.Zv[col] for col in cols]
# get a list of the views
V = list(set(views))
pK_dict = dict()
for v in V:
log_pK = get_cluster_crps(state, v)
pK = numpy.exp(utils.log_normalize(log_pK))
pK_dict[v] = pK
cluster_sets = dict()
for col in cols:
cluster_sets[col] = create_cluster_set(state, col)
draws = []
i = 0
view_dict = dict()
for v in V:
which_cols = numpy.nonzero(views == v)[0]
view_dict[v] = [col for col in which_cols]
for _ in range(N):
row_data = []
for v in V:
cols_v = view_dict[v]
k = utils.pflip(pK_dict[v])
for col in cols_v:
x = cluster_sets[col][k].predictive_draw()
row_data.append(x)
draws.append(row_data)
return draws
def _simple_predictive_sample_unobserved(state, cols, N=1):
# get if in the same view
views = [state.Zv[col] for col in cols]
# get a list of the views
V = list(set(views))
pK_dict = dict()
for v in V:
log_pK = get_cluster_crps(state, v)
pK = numpy.exp(utils.log_normalize(log_pK))
pK_dict[v] = pK
cluster_sets = dict()
for col in cols:
cluster_sets[col] = create_cluster_set(state, col)
draws = []
i = 0
view_dict = dict()
for v in V:
which_cols = numpy.nonzero(views == v)[0]
view_dict[v] = [col for col in which_cols]
for _ in range(N):
row_data = []
for v in V:
cols_v = view_dict[v]
k = utils.pflip(pK_dict[v])
for col in cols_v:
x = cluster_sets[col][k].predictive_draw()
row_data.append(x)
draws.append(row_data)
return draws
def predictive_draw(self):
return utils.pflip(w)
def predictive_draw(self):
return utils.pflip(w)