def _find_bcs():
classes = dict()
bcs = dict()
ucs = dict()
submods = dict()
for obname, ob in diter(pysal.spreg.__dict__):
if isinstance(ob, ty.ModuleType):
if ob.__package__.startswith("pysal"):
submods.update({obname: ob})
elif isinstance(ob, clstypes):
classes.update({obname: ob})
if ob.__name__.startswith('Base'):
bcs.update({obname: ob})
for modname, mod in diter(submods):
basecands = dict()
for clname in dir(mod):
cl = mod.__dict__[clname]
if isinstance(cl, clstypes):
try:
if cl.__name__.startswith('Base'):
if cl not in bcs:
bcs.update({cl.__name__: cl})
else:
classes.update({cl.__name__: cl})
except:
pass
ucs.update({
k: v
for k, v in diter(classes)
if (any([issubclass(v, bc) for bc in list(bcs.values())]) and
(k not in bcs)) or k.endswith('Regimes')
})
return bcs, ucs
def _find_bcs():
classes = dict()
bcs = dict()
ucs = dict()
submods = dict()
for obname, ob in diter(pysal.spreg.__dict__):
if isinstance(ob, ty.ModuleType):
if ob.__package__.startswith("pysal"):
submods.update({obname:ob})
elif isinstance(ob, clstypes):
classes.update({obname:ob})
if ob.__name__.startswith('Base'):
bcs.update({obname:ob})
for modname, mod in diter(submods):
basecands = dict()
for clname in dir(mod):
cl = mod.__dict__[clname]
if isinstance(cl, clstypes):
try:
if cl.__name__.startswith('Base'):
if cl not in bcs:
bcs.update({cl.__name__:cl})
else:
classes.update({cl.__name__:cl})
except:
pass
ucs.update({k:v for k,v in diter(classes) if (
any([issubclass(v, bc) for bc in bcs.values()])
and (k not in bcs))
or k.endswith('Regimes')})
return bcs, ucs
def lag_categorical(w, y, ties='tryself'):
"""
Spatial lag operator for categorical variables.
Constructs the most common categories of neighboring observations, weighted
by their weight strength.
Parameters
----------
w : W
PySAL spatial weightsobject
y : iterable
iterable collection of categories (either int or
string) with dimensionality conforming to w (see examples)
ties : str
string describing the method to use when resolving
ties. By default, the option is "tryself",
and the category of the focal observation
is included with its neighbors to try
and break a tie. If this does not resolve the tie,
a winner is chosen randomly. To just use random choice to
break ties, pass "random" instead.
Also supported are selecting the ``lowest'' class in the
sorted list of ties or the ``highest'' class.
Returns
-------
an (n x k) column vector containing the most common neighboring observation
Notes
-----
This works on any array where the number of unique elements along the column
axis is less than the number of elements in the array, for any dtype.
That means the routine should work on any dtype that np.unique() can
compare.
Examples
--------
Set up a 9x9 weights matrix describing a 3x3 regular lattice. Lag one list of
categorical variables with no ties.
>>> import pysal
>>> import numpy as np
>>> np.random.seed(12345)
>>> w = pysal.lat2W(3, 3)
>>> y = ['a','b','a','b','c','b','c','b','c']
>>> y_l = pysal.weights.spatial_lag.lag_categorical(w, y)
>>> np.array_equal(y_l, np.array(['b', 'a', 'b', 'c', 'b', 'c', 'b', 'c', 'b']))
True
Explicitly reshape y into a (9x1) array and calculate lag again
>>> yvect = np.array(y).reshape(9,1)
>>> yvect_l = pysal.weights.spatial_lag.lag_categorical(w,yvect)
>>> check = np.array( [ [i] for i in ['b', 'a', 'b', 'c', 'b', 'c', 'b', 'c', 'b']] )
>>> np.array_equal(yvect_l, check)
True
compute the lag of a 9x2 matrix of categories
>>> y2 = ['a', 'c', 'c', 'd', 'b', 'a', 'd', 'd', 'c']
>>> ym = np.vstack((y,y2)).T
>>> ym_lag = pysal.weights.spatial_lag.lag_categorical(w,ym)
>>> check = np.array([['b', 'b'], ['a', 'c'], ['b', 'c'], ['c', 'd'], ['b', 'd'], ['c', 'c'], ['b', 'd'], ['c', 'd'], ['b', 'b']])
>>> np.array_equal(check, ym_lag)
True
"""
if isinstance(y, list):
y = np.array(y)
orig_shape = y.shape
if len(orig_shape) > 1:
if orig_shape[1] > 1:
return np.vstack([lag_categorical(w, col) for col in y.T]).T
y = y.flatten()
output = np.zeros_like(y)
keys = np.unique(y)
inty = np.zeros(y.shape, dtype=np.int)
for i, key in enumerate(keys):
inty[y == key] = i
for idx, neighbors in w:
vals = np.zeros(keys.shape)
for neighb, weight in diter(neighbors):
vals[inty[w.id2i[neighb]]] += weight
outidx = _resolve_ties(idx, inty, vals, neighbors, ties, w)
output[w.id2i[int(idx)]] = keys[int(outidx)]
return output.reshape(orig_shape)
def lag_categorical(w, y, ties='tryself'):
"""
Spatial lag operator for categorical variables.
Constructs the most common categories of neighboring observations, weighted
by their weight strength.
Parameters
----------
w : W
PySAL spatial weightsobject
y : iterable
iterable collection of categories (either int or
string) with dimensionality conforming to w (see examples)
ties : str
string describing the method to use when resolving
ties. By default, the option is "tryself",
and the category of the focal observation
is included with its neighbors to try
and break a tie. If this does not resolve the tie,
a winner is chosen randomly. To just use random choice to
break ties, pass "random" instead.
Also supported are selecting the ``lowest'' class in the
sorted list of ties or the ``highest'' class.
Returns
-------
an (n x k) column vector containing the most common neighboring observation
Notes
-----
This works on any array where the number of unique elements along the column
axis is less than the number of elements in the array, for any dtype.
That means the routine should work on any dtype that np.unique() can
compare.
Examples
--------
Set up a 9x9 weights matrix describing a 3x3 regular lattice. Lag one list of
categorical variables with no ties.
>>> import pysal
>>> import numpy as np
>>> np.random.seed(12345)
>>> w = pysal.lat2W(3, 3)
>>> y = ['a','b','a','b','c','b','c','b','c']
>>> y_l = pysal.weights.spatial_lag.lag_categorical(w, y)
>>> np.array_equal(y_l, np.array(['b', 'a', 'b', 'c', 'b', 'c', 'b', 'c', 'b']))
True
Explicitly reshape y into a (9x1) array and calculate lag again
>>> yvect = np.array(y).reshape(9,1)
>>> yvect_l = pysal.weights.spatial_lag.lag_categorical(w,yvect)
>>> check = np.array( [ [i] for i in ['b', 'a', 'b', 'c', 'b', 'c', 'b', 'c', 'b']] )
>>> np.array_equal(yvect_l, check)
True
compute the lag of a 9x2 matrix of categories
>>> y2 = ['a', 'c', 'c', 'd', 'b', 'a', 'd', 'd', 'c']
>>> ym = np.vstack((y,y2)).T
>>> ym_lag = pysal.weights.spatial_lag.lag_categorical(w,ym)
>>> check = np.array([['b', 'b'], ['a', 'c'], ['b', 'c'], ['c', 'd'], ['b', 'd'], ['c', 'c'], ['b', 'd'], ['c', 'd'], ['b', 'b']])
>>> np.array_equal(check, ym_lag)
True
"""
if isinstance(y, list):
y = np.array(y)
orig_shape = y.shape
if len(orig_shape) > 1:
if orig_shape[1] > 1:
return np.vstack([lag_categorical(w,col) for col in y.T]).T
y = y.flatten()
output = np.zeros_like(y)
keys = np.unique(y)
inty = np.zeros(y.shape, dtype=np.int)
for i,key in enumerate(keys):
inty[y == key] = i
for idx,neighbors in w:
vals = np.zeros(keys.shape)
for neighb, weight in diter(neighbors):
vals[inty[w.id2i[neighb]]] += weight
outidx = _resolve_ties(idx,inty,vals,neighbors,ties, w)
output[w.id2i[int(idx)]] = keys[int(outidx)]
return output.reshape(orig_shape)
