def _get_partition_specs(self, uniqueattrs):
if isinstance(self.cvtype, float):
n = int(self.cvtype * len(uniqueattrs))
else:
n = self.cvtype
if self.count is None \
or self.selection_strategy != 'random' \
or self.count >= support.ncombinations(len(uniqueattrs), n):
# all combinations were requested so no need for
# randomization
combs = support.xunique_combinations(uniqueattrs, n)
else:
# due to selection_strategy=random they would be also
# reshuffled by super class later on but that should be ok
combs = support.xrandom_unique_combinations(
uniqueattrs, n, self.count)
if self.count is None or self.selection_strategy != 'random':
# we are doomed to return all of them
return [(None, i) for i in combs]
else:
# It makes sense to limit number of returned combinations
# right away
return [(None, i) for ind, i in enumerate(combs)
if ind < self.count]
def _get_partition_specs(self, uniqueattrs):
if isinstance(self.cvtype, float):
n = int(self.cvtype * len(uniqueattrs))
else:
n = self.cvtype
if self.count is None \
or self.selection_strategy != 'random' \
or self.count >= support.ncombinations(len(uniqueattrs), n):
# all combinations were requested so no need for
# randomization
combs = support.xunique_combinations(uniqueattrs, n)
else:
# due to selection_strategy=random they would be also
# reshuffled by super class later on but that should be ok
combs = support.xrandom_unique_combinations(uniqueattrs, n,
self.count)
if self.count is None or self.selection_strategy != 'random':
# we are doomed to return all of them
return [(None, i) for i in combs]
else:
# It makes sense to limit number of returned combinations
# right away
return [(None, i) for ind, i in enumerate(combs)
if ind < self.count]
def generate(self, ds):
orig_partitioning = ds.sa[self._partitions_attr].value.copy()
targets = ds.sa[self._targets_attr].value
testing_part = orig_partitioning == self._partitions_keep
nontesting_part = np.logical_not(testing_part)
utargets = np.unique(targets[testing_part])
for combination in xunique_combinations(utargets, self._k):
partitioning = orig_partitioning.copy()
combination_matches = [ t in combination for t in targets ]
combination_nonmatches = np.logical_not(combination_matches)
partitioning[np.logical_and(testing_part, combination_nonmatches)] = self._partition_assign
partitioning[np.logical_and(nontesting_part, combination_matches)] = self._partition_assign
pds = ds.copy(deep=False)
pds.sa[self.space] = partitioning
yield pds
def generate(self, ds):
orig_partitioning = ds.sa[self.partitions_attr].value.copy()
targets = ds.sa[self.targets_attr].value
testing_part = orig_partitioning == self.partitions_keep
nontesting_part = np.logical_not(testing_part)
utargets = np.unique(targets[testing_part])
for combination in support.xunique_combinations(utargets, self.k):
partitioning = orig_partitioning.copy()
combination_matches = [t in combination for t in targets]
combination_nonmatches = np.logical_not(combination_matches)
partitioning[np.logical_and(testing_part,
combination_nonmatches)] \
= self.partition_assign
partitioning[np.logical_and(nontesting_part,
combination_matches)] \
= self.partition_assign
pds = ds.copy(deep=False)
pds.sa[self.space] = partitioning
yield pds
def _get_partition_specs(self, uniqueattrs):
return [(None, i) for i in \
support.xunique_combinations(uniqueattrs, self.__cvtype)]
