def create_counter_types_matrix(locs, types, retriever_o, pars_ret):
"""Create network of types.
Parameters
----------
locs: array_like, shape (n, n_dim)
the spatial information of the retrievable elements.
types: array_like or list
the types codes.
retriever_o: pst.BaseRetriever
the retriever class (not instantiated yet).
pars_ret: dict
the parameters of the retriever.
Returns
-------
counts: np.ndarray or scipy.sparse
the counts matrix.
types_u: array_like or list
the types codes in the same order of the counts matrix.
"""
## 0. Prepare Retriever
ret = retriever_o(locs, **pars_ret)
counterdesc = CountDescriptor()
nulldesc = NullPhantomDescriptor()
## 1. Prepare variables
types_u = np.unique(types)
types_tr = np.array([np.where(e == types_u)[0][0] for e in types])
## 2. Prepare features
feats = ImplicitFeatures(types_tr, descriptormodel=counterdesc)
feats_ph = PhantomFeatures((len(locs), 1), descriptormodel=nulldesc)
## 3. Prepare builders
vals = np.arange(len(types))
feats_ret = FeaturesManager([feats, feats_ph],
selectors=[(0, 1), (0, 0), (1, 0)],
maps_vals_i=vals)
counter = SpatialDescriptorModel(ret, feats_ret)
## 4. Compute counter
counts = counter.compute()
return counts, types_u
centroids = (np.random.random((nclusters, 2))-.5)*10
selectors = ((0, 0), (0, 1), (0, 0))
avgdesc = AvgPosition()
avgdesc.set_global_info(np.arange(nclusters))
feats = ImplicitFeatures(np.arange(nclusters), descriptormodel=avgdesc)
## Performing algorithm
for i in range(n_steps):
## Expectation (assignation of the points)
ret0 = KRetriever(locs=centroids, autolocs=points, info_ret=nclusters,
ifdistance=True,
output_map=_output_map_mindist_filter)
feats_ret = FeaturesManager([points, feats],
maps_vals_i=np.zeros(len(points)),
selectors=selectors,
descriptormodels=AvgJoinerPosition())
spdesc = SpatialDescriptorModel(ret0, feats_ret)
measure = spdesc.compute()
#### Case of different axis variance
## Initialization of the points
# Parameters
nclusters = 3
# Random clustered points
x = np.concatenate([np.random.normal(5, 0.2, size=500),
np.random.normal(-4, 2.0, size=500),
np.random.normal(-3, 2.4, size=1000)])
y = np.concatenate([np.random.normal(4, 5, size=500),
np.random.normal(-3, 0.4, size=500),
np.random.normal(0, 0.2, size=1000)])
# Random initialization
x0 = (np.random.random(nclusters)-0.5)*10
