def _getadvanced(self, index):
"""
Advanced indexing (for sets, lists, or ndarrays).
"""
index = [asarray(i) for i in index]
shape = index[0].shape
if not all([i.shape == shape for i in index]):
raise ValueError(
"shape mismatch: indexing arrays could not be broadcast "
"together with shapes " +
("%s " * self.ndim) % tuple([i.shape for i in index]))
index = tuple([listify(i, d) for (i, d) in zip(index, self.shape)])
# build tuples with target indices
key_tuples = list(zip(*index[0:self.split]))
value_tuples = list(zip(*index[self.split:]))
# build dictionary to look up targets in values
d = {}
for k, g in groupby(zip(value_tuples, key_tuples), lambda x: x[1]):
d[k] = map(lambda x: x[0], list(g))
def key_check(key):
return key in key_tuples
def key_func(key):
return unravel_index(key, shape)
# filter records based on key targets
filtered = self._rdd.filter(lambda kv: key_check(kv[0]))
# subselect and flatten records based on value targets (if they exist)
if len(value_tuples) > 0:
flattened = filtered.flatMap(
lambda kv: [(kv[0], kv[1][i]) for i in d[kv[0]]])
else:
flattened = filtered
# reindex
indexed = flattened.zipWithIndex()
rdd = indexed.map(lambda kkv: (key_func(kkv[1]), kkv[0][1]))
split = len(shape)
return rdd, shape, split
def _getadvanced(self, index):
"""
Advanced indexing (for sets, lists, or ndarrays).
"""
index = [asarray(i) for i in index]
shape = index[0].shape
if not all([i.shape == shape for i in index]):
raise ValueError("shape mismatch: indexing arrays could not be broadcast "
"together with shapes " + ("%s " * self.ndim)
% tuple([i.shape for i in index]))
index = tuple([listify(i, d) for (i, d) in zip(index, self.shape)])
# build tuples with target indices
key_tuples = list(zip(*index[0:self.split]))
value_tuples = list(zip(*index[self.split:]))
# build dictionary to look up targets in values
d = {}
for k, g in groupby(zip(value_tuples, key_tuples), lambda x: x[1]):
d[k] = map(lambda x: x[0], list(g))
def key_check(key):
return key in key_tuples
def key_func(key):
return unravel_index(key, shape)
# filter records based on key targets
filtered = self._rdd.filter(lambda kv: key_check(kv[0]))
# subselect and flatten records based on value targets (if they exist)
if len(value_tuples) > 0:
flattened = filtered.flatMap(lambda kv: [(kv[0], kv[1][i]) for i in d[kv[0]]])
else:
flattened = filtered
# reindex
indexed = flattened.zipWithIndex()
rdd = indexed.map(lambda kkv: (key_func(kkv[1]), kkv[0][1]))
split = len(shape)
return rdd, shape, split
