def collapse_as(stepmon, offset=False, tolerance=0.005, \
generations=50, mask=None):
'''return a set of pairs of indices where the parameters exhibit a
dimensional collapse. Dimensional collapse is defined by:
max(pairwise(parameters)) <= tolerance over N generations (offset=False),
ptp(pairwise(parameters)) <= tolerance over N generations (offset=True).
collapse will be ignored at any pairs of indices specififed in the mask.
If single indices are provided, ignore all pairs with the given indices.
'''
np = _m.numpy
# reject bad masks
if mask is None: pass
elif type(mask) is set:
for i in mask:
if hasattr(i, '__len__') and len(i) != 2:
msg = "bad element '%s' in mask" % str(i)
raise ValueError(msg)
else:
msg = "%s is not a valid mask" % str(mask)
raise TypeError(msg)
# is the max position difference less than tolerance across all generations?
distances = _m._solutions(stepmon, generations)
#FIXME: HACK: array should be ndim=2... apparently sometimes it's ndim=3
if distances.ndim == 3 and distances.shape[-1] == 1:
distances.shape = distances.shape[:-1]
elif distances.ndim < 3:
pass
else:
msg = 'could not extract pairwise distances from array with shape %s' % distances.shape
raise ValueError(msg)
nindices = distances.shape[-1]
# get distances and pairs of indices
from mystic.tools import pairwise
distances, pairs = pairwise(distances, True)
if offset: # tracking at a distance
distances = distances.ptp(axis=0) <= tolerance
else: # tracking with the same position
distances = distances.max(axis=0) <= tolerance
# get the (index1,index2) pairs where the collapse occurs
if distances.ndim > 1:
distances.shape = tuple(i for i in distances.shape if i != 1) or (1, )
distances = np.array(pairs)[distances]
# apply mask
if mask is None: return set(tuple(i) for i in distances)
mask = selector(mask)
return set(tuple(i) for i in distances if not mask(i))
def collapse_as(stepmon, offset=False, tolerance=0.005, \
generations=50, mask=None):
'''return a set of pairs of indices where the parameters exhibit a
dimensional collapse. Dimensional collapse is defined by:
max(pairwise(parameters)) <= tolerance over N generations (offset=False),
ptp(pairwise(parameters)) <= tolerance over N generations (offset=True).
collapse will be ignored at any pairs of indices specififed in the mask.
If single indices are provided, ignore all pairs with the given indices.
'''
np = _m.numpy
# reject bad masks
if mask is None: pass
elif type(mask) is set:
for i in mask:
if hasattr(i, '__len__') and len(i) != 2:
msg = "bad element '%s' in mask" % str(i)
raise ValueError(msg)
else:
msg = "%s is not a valid mask" % str(mask)
raise TypeError(msg)
# is the max position difference less than tolerance across all generations?
distances = _m._solutions(stepmon, generations)
#FIXME: HACK: array should be ndim=2... apparently sometimes it's ndim=3
if distances.ndim == 3 and distances.shape[-1] == 1:
distances.shape = distances.shape[:-1]
elif distances.ndim < 3:
pass
else:
msg = 'could not extract pairwise distances from array with shape %s' % distances.shape
raise ValueError(msg)
nindices = distances.shape[-1]
# get distances and pairs of indices
from mystic.tools import pairwise
distances, pairs = pairwise(distances, True)
if offset: # tracking at a distance
distances = distances.ptp(axis=0) <= tolerance
else: # tracking with the same position
distances = distances.max(axis=0) <= tolerance
# get the (index1,index2) pairs where the collapse occurs
if distances.ndim > 1:
distances.shape = tuple(i for i in distances.shape if i != 1) or (1,)
distances = np.array(pairs)[distances]
# apply mask
if mask is None: return set(tuple(i) for i in distances)
mask = selector(mask)
return set(tuple(i) for i in distances if not mask(i))
def collapse_position(stepmon, tolerance=0.005, generations=50, mask=None):
'''return a dict of {measure: pairs_of_indices} where the product_measure
exhibits a dimensional collapse in position. Dimensional collapse in
position is defined by:
collapse will be ignored at (measure,pairs) as specified in the mask.
Format of mask will determine the return value for this function. Default
mask format is dict of {measure: pairs_of_indices}, with alternate
formatting available as a set of tuples of (measure,pair).
''' #XXX: not mentioned, 'where' format also available
np = _m.numpy
# reject bad masks
if mask is None: pass
elif type(mask) is set:
for i in mask:
if not hasattr(i, '__len__') or len(i) != 2:
msg = "bad element '%s' in mask" % str(i)
raise ValueError(msg)
if type(i[0]) is not int:
msg = "bad element '%s' in mask" % str(i)
raise ValueError(msg)
if np.array(i[1]).ndim != 1:
msg = "bad element '%s' in mask" % str(i)
raise ValueError(msg)
elif type(mask) is dict:
for (i, j) in getattr(mask, 'iteritems', mask.items)():
if type(j) is not set or type(i) is not int:
msg = "bad entry '%s:%s' in mask" % (str(i), str(j))
raise ValueError(msg)
for k in j: # items in the set
if not hasattr(k, '__len__') or len(k) != 2:
msg = "bad entry '%s:%s' in mask" % (str(i), str(j))
raise ValueError(msg)
elif hasattr(mask, '__len__') and len(mask) == 2:
if np.array(mask[0]).ndim != 1:
msg = "%s is not a valid mask" % str(mask)
raise TypeError(msg)
if np.array(mask[1]).ndim != 2:
msg = "%s is not a valid mask" % str(mask)
raise TypeError(msg)
elif hasattr(mask, '__len__') and not len(mask):
mask = type(mask)(((), ())) #XXX: HACK to get empty where mask
else:
msg = "%s is not a valid mask" % str(mask)
raise TypeError(msg)
# is the max position difference less than tolerance across all generations?
distances = _m._positions(stepmon, generations)
nindices = distances.shape[-1]
from mystic.tools import pairwise
distances, pairs = pairwise(distances, True)
distances = distances.max(axis=0) <= tolerance
# select off the desired pairs (of indices)
counts = np.cumsum(distances.sum(axis=-1))
import warnings
with warnings.catch_warnings(): #FIXME: python2.5
warnings.simplefilter('ignore')
#XXX: throws a FutureWarning
distances = np.split(
np.array((pairs, ) * distances.shape[0])[distances], counts)[:-1]
counts = (set(tuple(i) for i in j) for j in distances)
# return in terms of pairs, b/c indexing alone doesn't give pairings
# (keys are measure, and values are (index1,index2) pairs)
distances = ((i, j) for (i, j) in enumerate(counts) if len(j))
# identify mask format and build filter
select, pairs = _position_filter(mask)
# convert to selected format...
if pairs: # return explicit 'pairs' {(measure,indices)}
import itertools
mask = set()
for i, j in distances:
[
mask.add(k)
for k in getattr(itertools, 'izip', zip)(*((i, ) * len(j), j))
]
elif pairs is None: # return 'where' format [measures,indices]
import itertools
# tuple of where,pairs
measures, mask = (), ()
for (i, j) in (getattr(itertools, 'izip', zip)(*((j[0], i)
for i in j[1]))
for j in distances):
measures += i
mask += j
mask = (measures, mask) if len(measures) else ()
else:
# returns a dict of {measure, indices}
mask = dict(distances)
# apply mask
return select(mask)
def collapse_position(stepmon, tolerance=0.005, generations=50, mask=None):
'''return a dict of {measure: pairs_of_indices} where the product_measure
exhibits a dimensional collapse in position. Dimensional collapse in
position is defined by:
collapse will be ignored at (measure,pairs) as specified in the mask.
Format of mask will determine the return value for this function. Default
mask format is dict of {measure: pairs_of_indices}, with alternate
formatting available as a set of tuples of (measure,pair).
''' #XXX: not mentioned, 'where' format also available
np = _m.numpy
# reject bad masks
if mask is None: pass
elif type(mask) is set:
for i in mask:
if not hasattr(i, '__len__') or len(i) != 2:
msg = "bad element '%s' in mask" % str(i)
raise ValueError(msg)
if type(i[0]) is not int:
msg = "bad element '%s' in mask" % str(i)
raise ValueError(msg)
if np.array(i[1]).ndim != 1:
msg = "bad element '%s' in mask" % str(i)
raise ValueError(msg)
elif type(mask) is dict:
for (i,j) in getattr(mask, 'iteritems', mask.items)():
if type(j) is not set or type(i) is not int:
msg = "bad entry '%s:%s' in mask" % (str(i),str(j))
raise ValueError(msg)
for k in j: # items in the set
if not hasattr(k, '__len__') or len(k) != 2:
msg = "bad entry '%s:%s' in mask" % (str(i),str(j))
raise ValueError(msg)
elif hasattr(mask, '__len__') and len(mask) == 2:
if np.array(mask[0]).ndim != 1:
msg = "%s is not a valid mask" % str(mask)
raise TypeError(msg)
if np.array(mask[1]).ndim != 2:
msg = "%s is not a valid mask" % str(mask)
raise TypeError(msg)
elif hasattr(mask, '__len__') and not len(mask):
mask = type(mask)(((),())) #XXX: HACK to get empty where mask
else:
msg = "%s is not a valid mask" % str(mask)
raise TypeError(msg)
# is the max position difference less than tolerance across all generations?
distances = _m._positions(stepmon, generations)
nindices = distances.shape[-1]
from mystic.tools import pairwise
distances, pairs = pairwise(distances, True)
distances = distances.max(axis=0) <= tolerance
# select off the desired pairs (of indices)
counts = np.cumsum(distances.sum(axis=-1))
import warnings
with warnings.catch_warnings(): #FIXME: python2.5
warnings.simplefilter('ignore')
#XXX: throws a FutureWarning
distances = np.split(np.array((pairs,)*distances.shape[0])[distances], counts)[:-1]
counts = (set(tuple(i) for i in j) for j in distances)
# return in terms of pairs, b/c indexing alone doesn't give pairings
# (keys are measure, and values are (index1,index2) pairs)
distances = ((i,j) for (i,j) in enumerate(counts) if len(j))
# identify mask format and build filter
select, pairs = _position_filter(mask)
# convert to selected format...
if pairs: # return explicit 'pairs' {(measure,indices)}
import itertools
mask = set()
for i,j in distances:
[mask.add(k) for k in getattr(itertools, 'izip', zip)(*((i,)*len(j), j))]
elif pairs is None: # return 'where' format [measures,indices]
import itertools
# tuple of where,pairs
measures,mask = (),()
for (i,j) in (getattr(itertools, 'izip', zip)(*((j[0],i) for i in j[1])) for j in distances):
measures += i
mask += j
mask = (measures,mask) if len(measures) else ()
else:
# returns a dict of {measure, indices}
mask = dict(distances)
# apply mask
return select(mask)