def for_intra(dpack, target):
"""Adapt a datapack to intrasentential decoding.
An intrasentential datapack is almost identical to its original,
except that we set the label for each ('ROOT', edu) pairing to
'ROOT' if that edu is a subgrouping head (if it has no parents other
than 'ROOT' within its subgrouping).
This should be done before either `for_labelling` or `for_attachment`
Returns
-------
dpack : DataPack
target : array(int)
"""
# map EDUs to subgroup ids ; intra = pairs of EDUs with same subgroup id
grp = {e.id: e.subgrouping for e in dpack.edus}
# find all edus that have intra incoming edges (to rule out)
unrelated = dpack.label_number(UNRELATED)
intra_tgts = defaultdict(set)
for i, (edu1, edu2) in enumerate(dpack.pairings):
if (grp[edu1.id] == grp[edu2.id]
and target[i] != unrelated):
# edu2 has an incoming relation => not an (intra) root
intra_tgts[grp[edu2.id]].add(edu2.id)
# pick out the (fakeroot, edu) pairs where edu does not have
# incoming intra edges
all_heads = [i for i, (edu1, edu2) in enumerate(dpack.pairings)
if (edu1.id == FAKE_ROOT_ID
and edu2.id not in intra_tgts[grp[edu2.id]])]
# NEW pick out the original inter-sentential links, for removal
inter_links = [i for i, (edu1, edu2) in enumerate(dpack.pairings)
if (edu1.id != FAKE_ROOT_ID
and grp[edu1.id] != grp[edu2.id]
and target[i] != unrelated)]
# update datapack and target accordingly
new_target = np.copy(dpack.target)
new_target[all_heads] = dpack.label_number('ROOT')
new_target[inter_links] = unrelated # NEW
# WIP ctarget
new_ctarget = {grp_name: ctgt
for grp_name, ctgt in dpack.ctarget.items()}
# FIXME replace each ctgt with the list of intra-sentential
# RST (sub)trees
# end WIP ctarget
dpack = DataPack(edus=dpack.edus,
pairings=dpack.pairings,
data=dpack.data,
target=new_target,
ctarget=new_ctarget,
labels=dpack.labels,
vocab=dpack.vocab,
graph=dpack.graph)
target = np.copy(target)
target[all_heads] = dpack.label_number('ROOT')
target[inter_links] = unrelated # NEW
return dpack, target
def for_intra(dpack, target):
"""Adapt a datapack to intrasentential decoding.
An intrasenential datapack is almost identical to its original, except that
we set the label for each ('ROOT', edu) pairing to 'ROOT' if that edu is a
subgrouping head (if it has no parents other than than 'ROOT' within its
subgrouping).
This should be done before either `for_labelling` or `for_attachment`
Returns
-------
dpack: DataPack
target: array(int)
"""
# find all edus that have intra incoming edges (to rule out)
unrelated = dpack.label_number(UNRELATED)
intra_tgts = defaultdict(set)
for i, (edu1, edu2) in enumerate(dpack.pairings):
subg = edu2.subgrouping
if (edu1.subgrouping == subg and
target[i] != unrelated):
intra_tgts[subg].add(edu2.id)
# pick out the (fakeroot, edu) pairs where edu does not have
# incoming intra edges
all_heads = []
for i, (edu1, edu2) in enumerate(dpack.pairings):
subg = edu2.subgrouping
if (edu1.id == FAKE_ROOT_ID and
edu2.id not in intra_tgts[subg]):
all_heads.append(i)
# update datapack and target accordingly
new_target = np.copy(dpack.target)
new_target[all_heads] = dpack.label_number('ROOT')
dpack = DataPack(edus=dpack.edus,
pairings=dpack.pairings,
data=dpack.data,
target=new_target,
labels=dpack.labels,
vocab=dpack.vocab,
graph=dpack.graph)
target = np.copy(target)
target[all_heads] = dpack.label_number('ROOT')
return dpack, target
def for_intra(dpack, target):
"""Adapt a datapack to intrasentential decoding.
An intrasentential datapack is almost identical to its original,
except that we set the label for each ('ROOT', edu) pairing to
'ROOT' if that edu is a subgrouping head (if it has no parents other
than 'ROOT' within its subgrouping).
This should be done before either `for_labelling` or `for_attachment`
Returns
-------
dpack : DataPack
target : array(int)
"""
# map EDUs to subgroup ids ; intra = pairs of EDUs with same subgroup id
grp = {e.id: e.subgrouping for e in dpack.edus}
# find all edus that have intra incoming edges (to rule out)
unrelated = dpack.label_number(UNRELATED)
intra_tgts = defaultdict(set)
for i, (edu1, edu2) in enumerate(dpack.pairings):
if (grp[edu1.id] == grp[edu2.id] and target[i] != unrelated):
# edu2 has an incoming relation => not an (intra) root
intra_tgts[grp[edu2.id]].add(edu2.id)
# pick out the (fakeroot, edu) pairs where edu does not have
# incoming intra edges
all_heads = [
i for i, (edu1, edu2) in enumerate(dpack.pairings)
if (edu1.id == FAKE_ROOT_ID and edu2.id not in intra_tgts[grp[edu2.id]]
)
]
# NEW pick out the original inter-sentential links, for removal
inter_links = [
i for i, (edu1, edu2) in enumerate(dpack.pairings)
if (edu1.id != FAKE_ROOT_ID and grp[edu1.id] != grp[edu2.id]
and target[i] != unrelated)
]
# update datapack and target accordingly
new_target = np.copy(dpack.target)
new_target[all_heads] = dpack.label_number('ROOT')
new_target[inter_links] = unrelated # NEW
# WIP ctarget
new_ctarget = {grp_name: ctgt for grp_name, ctgt in dpack.ctarget.items()}
# FIXME replace each ctgt with the list of intra-sentential
# RST (sub)trees
# end WIP ctarget
dpack = DataPack(edus=dpack.edus,
pairings=dpack.pairings,
data=dpack.data,
target=new_target,
ctarget=new_ctarget,
labels=dpack.labels,
vocab=dpack.vocab,
graph=dpack.graph)
target = np.copy(target)
target[all_heads] = dpack.label_number('ROOT')
target[inter_links] = unrelated # NEW
return dpack, target