WIP on add_label bug during NER training

Currently when a new label is introduced to NER during training,
it causes the labels to be read in in an unexpected order. This
invalidates the model.
This commit is contained in:
Matthew Honnibal 2017-04-14 23:52:17 +02:00
parent 33ba5066eb
commit 354458484c
5 changed files with 65 additions and 44 deletions

View File

@ -17,17 +17,13 @@ cdef class EntityRecognizer(Parser):
feature_templates = get_feature_templates('ner')
def add_label(self, label):
for action in self.moves.action_types:
self.moves.add_action(action, label)
if 'actions' in self.cfg:
self.cfg['actions'].setdefault(action,
{}).setdefault(label, True)
Parser.add_label(self, label)
if isinstance(label, basestring):
label = self.vocab.strings[label]
# Set label into serializer. Super hacky :(
for attr, freqs in self.vocab.serializer_freqs:
if attr == ENT_TYPE and label not in freqs:
freqs.append([label, 1])
# Super hacky :(
self.vocab._serializer = None
@ -36,19 +32,15 @@ cdef class BeamEntityRecognizer(BeamParser):
TransitionSystem = BiluoPushDown
feature_templates = get_feature_templates('ner')
def add_label(self, label):
for action in self.moves.action_types:
self.moves.add_action(action, label)
if 'actions' in self.cfg:
self.cfg['actions'].setdefault(action,
{}).setdefault(label, True)
Parser.add_label(self, label)
if isinstance(label, basestring):
label = self.vocab.strings[label]
# Set label into serializer. Super hacky :(
for attr, freqs in self.vocab.serializer_freqs:
if attr == ENT_TYPE and label not in freqs:
freqs.append([label, 1])
# Super hacky :(
self.vocab._serializer = None
@ -58,11 +50,7 @@ cdef class DependencyParser(Parser):
feature_templates = get_feature_templates('basic')
def add_label(self, label):
for action in self.moves.action_types:
self.moves.add_action(action, label)
if 'actions' in self.cfg:
self.cfg['actions'].setdefault(action,
{}).setdefault(label, True)
Parser.add_label(self, label)
if isinstance(label, basestring):
label = self.vocab.strings[label]
for attr, freqs in self.vocab.serializer_freqs:
@ -78,11 +66,7 @@ cdef class BeamDependencyParser(BeamParser):
feature_templates = get_feature_templates('basic')
def add_label(self, label):
for action in self.moves.action_types:
self.moves.add_action(action, label)
if 'actions' in self.cfg:
self.cfg['actions'].setdefault(action,
{}).setdefault(label, True)
Parser.add_label(self, label)
if isinstance(label, basestring):
label = self.vocab.strings[label]
for attr, freqs in self.vocab.serializer_freqs:

View File

@ -317,17 +317,20 @@ cdef class ArcEager(TransitionSystem):
def get_actions(cls, **kwargs):
actions = kwargs.get('actions',
{
SHIFT: {'': True},
REDUCE: {'': True},
RIGHT: {},
LEFT: {},
BREAK: {'ROOT': True}})
SHIFT: [''],
REDUCE: [''],
RIGHT: [],
LEFT: [],
BREAK: ['ROOT']})
seen_actions = set()
for label in kwargs.get('left_labels', []):
if label.upper() != 'ROOT':
actions[LEFT][label] = True
if (LEFT, label) not in seen_actions:
actions[LEFT].append(label)
for label in kwargs.get('right_labels', []):
if label.upper() != 'ROOT':
actions[RIGHT][label] = True
if (RIGHT, label) not in seen_actions:
actions[RIGHT].append(label)
for raw_text, sents in kwargs.get('gold_parses', []):
for (ids, words, tags, heads, labels, iob), ctnts in sents:
@ -336,9 +339,11 @@ cdef class ArcEager(TransitionSystem):
label = 'ROOT'
if label != 'ROOT':
if head < child:
actions[RIGHT][label] = True
if (RIGHT, label) not in seen_actions:
actions[RIGHT].append(label)
elif head > child:
actions[LEFT][label] = True
if (LEFT, label) not in seen_actions:
actions[LEFT].append(label)
return actions
property action_types:

View File

@ -21,6 +21,7 @@ cdef enum:
LAST
UNIT
OUT
ISNT
N_MOVES
@ -31,6 +32,7 @@ MOVE_NAMES[IN] = 'I'
MOVE_NAMES[LAST] = 'L'
MOVE_NAMES[UNIT] = 'U'
MOVE_NAMES[OUT] = 'O'
MOVE_NAMES[ISNT] = 'x'
cdef do_func_t[N_MOVES] do_funcs
@ -54,16 +56,20 @@ cdef class BiluoPushDown(TransitionSystem):
def get_actions(cls, **kwargs):
actions = kwargs.get('actions',
{
MISSING: {'': True},
BEGIN: {},
IN: {},
LAST: {},
UNIT: {},
OUT: {'': True}
MISSING: [''],
BEGIN: [],
IN: [],
LAST: [],
UNIT: [],
OUT: ['']
})
seen_entities = set()
for entity_type in kwargs.get('entity_types', []):
if entity_type in seen_entities:
continue
seen_entities.add(entity_type)
for action in (BEGIN, IN, LAST, UNIT):
actions[action][entity_type] = True
actions[action].append(entity_type)
moves = ('M', 'B', 'I', 'L', 'U')
for raw_text, sents in kwargs.get('gold_parses', []):
for (ids, words, tags, heads, labels, biluo), _ in sents:
@ -72,8 +78,10 @@ cdef class BiluoPushDown(TransitionSystem):
if ner_tag.count('-') != 1:
raise ValueError(ner_tag)
_, label = ner_tag.split('-')
for move_str in ('B', 'I', 'L', 'U'):
actions[moves.index(move_str)][label] = True
if label not in seen_entities:
seen_entities.add(label)
for move_str in ('B', 'I', 'L', 'U'):
actions[moves.index(move_str)].append(label)
return actions
property action_types:
@ -111,11 +119,17 @@ cdef class BiluoPushDown(TransitionSystem):
label = 0
elif '-' in name:
move_str, label_str = name.split('-', 1)
# Hacky way to denote 'not this entity'
if label_str.startswith('!'):
label_str = label_str[1:]
move_str = 'x'
label = self.strings[label_str]
else:
move_str = name
label = 0
move = MOVE_NAMES.index(move_str)
if move == ISNT:
return Transition(clas=0, move=ISNT, label=label, score=0)
for i in range(self.n_moves):
if self.c[i].move == move and self.c[i].label == label:
return self.c[i]
@ -225,6 +239,9 @@ cdef class Begin:
elif g_act == BEGIN:
# B, Gold B --> Label match
return label != g_tag
# Support partial supervision in the form of "not this label"
elif g_act == ISNT:
return label == g_tag
else:
# B, Gold I --> False (P)
# B, Gold L --> False (P)
@ -359,6 +376,9 @@ cdef class Unit:
elif g_act == UNIT:
# U, Gold U --> True iff tag match
return label != g_tag
# Support partial supervision in the form of "not this label"
elif g_act == ISNT:
return label == g_tag
else:
# U, Gold B --> False
# U, Gold I --> False
@ -388,7 +408,7 @@ cdef class Out:
cdef int g_act = gold.ner[s.B(0)].move
cdef int g_tag = gold.ner[s.B(0)].label
if g_act == MISSING:
if g_act == MISSING or g_act == ISNT:
return 0
elif g_act == BEGIN:
# O, Gold B --> False

View File

@ -52,7 +52,7 @@ from ._parse_features cimport fill_context
from .stateclass cimport StateClass
from ._state cimport StateC
USE_FTRL = True
USE_FTRL = False
DEBUG = False
def set_debug(val):
global DEBUG
@ -152,6 +152,13 @@ cdef class Parser:
# TODO: remove this shim when we don't have to support older data
if 'labels' in cfg and 'actions' not in cfg:
cfg['actions'] = cfg.pop('labels')
# TODO: remove this shim when we don't have to support older data
for action_name, labels in dict(cfg['actions']).items():
# We need this to be sorted
if isinstance(labels, dict):
labels = list(sorted(labels.keys()))
cfg['actions'][action_name] = labels
print(cfg['actions'])
self = cls(vocab, TransitionSystem=TransitionSystem, model=None, **cfg)
if (path / 'model').exists():
self.model.load(str(path / 'model'))
@ -362,6 +369,10 @@ cdef class Parser:
# Doesn't set label into serializer -- subclasses override it to do that.
for action in self.moves.action_types:
self.moves.add_action(action, label)
if 'actions' in self.cfg:
# Important that the labels be stored as a list! We need the
# order, or the model goes out of synch
self.cfg['actions'].setdefault(str(action), []).append(label)
cdef class StepwiseState:

View File

@ -32,7 +32,7 @@ cdef class TransitionSystem:
self.c = <Transition*>self.mem.alloc(self._size, sizeof(Transition))
for action, label_strs in sorted(labels_by_action.items()):
for label_str in sorted(label_strs):
for label_str in label_strs:
self.add_action(int(action), label_str)
self.root_label = self.strings['ROOT']
self.freqs = {} if _freqs is None else _freqs
@ -105,5 +105,6 @@ cdef class TransitionSystem:
self.c = <Transition*>self.mem.realloc(self.c, self._size * sizeof(self.c[0]))
self.c[self.n_moves] = self.init_transition(self.n_moves, action, label)
print("Add action", action, self.strings[label], self.n_moves)
self.n_moves += 1
return 1