def get_pair_mentions_features(self, m1, m2):
''' Features for pair of mentions (same speakers, speaker mentioned, string match)'''
features_ = {"00_SameSpeaker": 1 if self.consider_speakers and m1.speaker == m2.speaker else 0,
"01_AntMatchMentionSpeaker": 1 if self.consider_speakers and m2.speaker_match_mention(m1) else 0,
"02_MentionMatchSpeaker": 1 if self.consider_speakers and m1.speaker_match_mention(m2) else 0,
"03_HeadsAgree": 1 if m1.heads_agree(m2) else 0,
"04_ExactStringMatch": 1 if m1.exact_match(m2) else 0,
"05_RelaxedStringMatch": 1 if m1.relaxed_match(m2) else 0,
"06_SentenceDistance": m2.utterances_sent - m1.utterances_sent,
"07_MentionDistance": m2.index - m1.index - 1,
"08_Overlapping": 1 if (m1.utterances_sent == m2.utterances_sent and m1.end > m2.start) else 0,
"09_M1Features": m1.features_,
"10_M2Features": m2.features_,
"11_DocGenre": self.genre_}
pairwise_features = [np.array([features_["00_SameSpeaker"],
features_["01_AntMatchMentionSpeaker"],
features_["02_MentionMatchSpeaker"],
features_["03_HeadsAgree"],
features_["04_ExactStringMatch"],
features_["05_RelaxedStringMatch"]]),
encode_distance(features_["06_SentenceDistance"]),
encode_distance(features_["07_MentionDistance"]),
np.array(features_["08_Overlapping"], ndmin=1),
m1.features,
m2.features,
self.genre]
return (features_, np.concatenate(pairwise_features, axis=0))
def set_mentions_features(self):
'''
Compute features for the extracted mentions
'''
doc_embedding = self.embed_extractor.get_document_embedding(self.utterances) if self.embed_extractor is not None else None
for mention in self.mentions:
one_hot_type = np.zeros((4,))
one_hot_type[mention.mention_type] = 1
features_ = {"01_MentionType": mention.mention_type,
"02_MentionLength": len(mention)-1,
"03_MentionNormLocation": (mention.index)/len(self.mentions),
"04_IsMentionNested": 1 if any((m is not mention
and m.utterances_sent == mention.utterances_sent
and m.start <= mention.start
and mention.end <= m.end)
for m in self.mentions) else 0}
features = np.concatenate([one_hot_type,
encode_distance(features_["02_MentionLength"]),
np.array(features_["03_MentionNormLocation"], ndmin=1, copy=False),
np.array(features_["04_IsMentionNested"], ndmin=1, copy=False)
], axis=0)
(spans_embeddings_, words_embeddings_,
spans_embeddings, words_embeddings) = self.embed_extractor.get_mention_embeddings(mention, doc_embedding)
mention.features_ = features_
mention.features = features
mention.spans_embeddings = spans_embeddings
mention.spans_embeddings_ = spans_embeddings_
mention.words_embeddings = words_embeddings
mention.words_embeddings_ = words_embeddings_
def get_pair_mentions_features(self, m1, m2, doc_id, fileCsv, fileId,
fileLabel, fileFeature, setID):
''' Features for pair of mentions (same speakers, speaker mentioned, string match)'''
features_ = {
"00_SameSpeaker":
1 if self.consider_speakers and m1.speaker == m2.speaker else 0,
"01_AntMatchMentionSpeaker":
1
if self.consider_speakers and m2.speaker_match_mention(m1) else 0,
"02_MentionMatchSpeaker":
1
if self.consider_speakers and m1.speaker_match_mention(m2) else 0,
"03_HeadsAgree":
1 if m1.heads_agree(m2) else 0,
"04_ExactStringMatch":
1 if m1.exact_match(m2) else 0,
"05_RelaxedStringMatch":
1 if m1.relaxed_match(m2) else 0,
"06_SentenceDistance":
m2.utterances_sent - m1.utterances_sent,
"07_MentionDistance":
m2.index - m1.index - 1,
"08_Overlapping":
1 if (m1.utterances_sent == m2.utterances_sent
and m1.end > m2.start) else 0,
"09_M1Features":
m1.features_,
"10_M2Features":
m2.features_,
"11_DocGenre":
self.genre_
}
pairwise_features = [
np.array([
features_["00_SameSpeaker"],
features_["01_AntMatchMentionSpeaker"],
features_["02_MentionMatchSpeaker"],
features_["03_HeadsAgree"], features_["04_ExactStringMatch"],
features_["05_RelaxedStringMatch"]
]),
encode_distance(features_["06_SentenceDistance"]),
encode_distance(features_["07_MentionDistance"]),
np.array(features_["08_Overlapping"], ndmin=1), m1.features,
m2.features, self.genre
]
folderIndexMention = "/home/hung/git/neuralcoref/unite/index/"
#folderFeatureMention="/home/hung/git/neuralcoref/feature_mentions/"
strFileM1 = "".join(
[folderIndexMention,
str(doc_id), "_",
str(m1.index), ".txt"])
strFileM2 = "".join(
[folderIndexMention,
str(doc_id), "_",
str(m2.index), ".txt"])
#strFileFeat="".join([folderFeatureMention,str(m1.index),"_",str(m2.index),".txt"])
if not os.path.isfile(strFileM1):
strWriteM1 = u"".join([
str(doc_id), "\n",
str(m1.index), "\n",
str(m1.gold_label), "\n",
str(m1.start), "\n",
str(m1.end), "\n",
str(m1).encode('utf8'), "\n"
])
text_file = open(strFileM1, "w")
text_file.write(strWriteM1)
text_file.close()
if not os.path.isfile(strFileM2):
strWriteM2 = u"".join([
str(doc_id), "\n",
str(m2.index), "\n",
str(m2.gold_label), "\n",
str(m2.start), "\n",
str(m2.end), "\n",
str(m2).encode('utf8'), "\n"
])
text_file = open(strFileM2, "w")
text_file.write(strWriteM2)
text_file.close()
finalResult = np.concatenate(pairwise_features, axis=0)
strId = "".join(
[str(doc_id), "_",
str(m1.index), "_",
str(m2.index), ".txt"]).replace("\n", " ").strip()
if strId not in setID:
setID.add(strId)
strPairLabel = "0"
if m1.gold_label == m2.gold_label:
strPairLabel = "1"
#strWriteFeat=u"".join([str(m1.index),"\n",str(m1.gold_label),"\n",str(m2.index),"\n",str(m2.gold_label),"\n",strPairLabel,"\n",str(finalResult).replace("\n", " "),"\n"])
strFeat = "".join([str(doc_id), ",", strId, ","])
for i in range(len(finalResult)):
strFeat = "".join([strFeat, str(finalResult[i]), ","])
strFeat = "".join([strFeat, strPairLabel])
#strFeat=str(finalResult).replace("\n", " ")
fileId.write("".join([strId, "\n"]))
fileLabel.write("".join([strPairLabel, "\n"]))
fileFeature.write("".join([strFeat, "\n"]))
fileCsv.write("".join([strFeat, "\n"]))
#print(m1.gold_label+" "+m2.gold_label+" "+strPairLabel)
#print('m1 '+' '+str(m1)+' '+str(m1.start)+' '+str(m1.end)+' '+str(m1.index)+' '+str(m1.gold_label)+' abc')
#print('m2 '+' '+str(m2)+' '+str(m2.start)+' '+str(m2.end)+' '+str(m2.index)+' '+str(m2.gold_label)+' abc')
#print('pair'+str(np.concatenate(pairwise_features, axis=0)))
#abc = raw_input("What is your name? ")
return (features_, finalResult)
def __getitem__(self, mention_idx, debug=False):
"""
Return:
Definitions:
P is the number of antecedent per mention (number of pairs for the mention)
S = 250 is the size of the span vector (averaged word embeddings)
W = 8 is the number of words in a mention (tuned embeddings)
Fp = 70 is the number of features for a pair of mention
Fs = 24 is the number of features of a single mention
if there are some pairs:
inputs = (spans, words, features, ant_spans, ant_words, ana_spans, ana_words, pairs_features)
targets = (labels, costs, true_ants, false_ants)
else:
inputs = (spans, words, features)
targets = (labels, costs, true_ants)
inputs: Tuple of
spans => (S,) 250
words => (W,) 8
features => (Fs,) 70
+ if there are potential antecedents (P > 0):
ant_spans => (P, S) or nothing if no pairs (P,250)
ant_words => (P, W) or nothing if no pairs (P,8)
ana_spans => (P, S) or nothing if no pairs
ana_words => (P, W) or nothing if no pairs
pair_features => (P, Fp) or nothing if no pairs
targets: Tuple of
labels => (P+1,)
costs => (P+1,)
true_ant => (P+1,)
+ if there are potential antecedents (P > 0):
false_ant => (P+1,)
"""
features_raw, label, pairs_length, pairs_start_index, spans, words = self.mentions[
mention_idx]
pairs_start_index = np.asscalar(pairs_start_index)
pairs_length = np.asscalar(pairs_length)
# Build features array (float) from raw features (int)
#Fs共24个特征
### SIZE_FS_COMPRESSED size of the features for a mention as stored in numpy arrays分割成4,11,1,1，
assert features_raw.shape[0] == SIZE_FS_COMPRESSED
##24*
features = np.zeros((SIZE_FS, ))
features[features_raw[0]] = 1
##features_raw[1]进行one-hot
features[4:15] = encode_distance(features_raw[1])
features[15] = features_raw[2].astype(float) / features_raw[3].astype(
float)
features[16] = features_raw[4]
features[features_raw[5] + 17] = 1
if pairs_length == 0:
spans = torch.from_numpy(spans).float()
words = torch.from_numpy(words)
features = torch.from_numpy(features).float()
inputs = (spans, words, features)
if self.no_targets:
return inputs
true_ant = torch.zeros(1).long() # zeros = indices of true ant
costs = torch.from_numpy((1 - label) * self.costs['FN']).float()
label = torch.from_numpy(label).float()
targets = (label, costs, true_ant)
if debug:
print("inputs shapes: ", [a.size() for a in inputs])
print("targets shapes: ", [a.size() for a in targets])
return inputs, targets
##ant_spans, ant_words, ana_spans, ana_words, pairs_features
start = pairs_start_index
end = pairs_start_index + pairs_length
pairs = self.pairs[start:end]
assert len(pairs) == pairs_length
assert len(
pairs[0]
) == 3 # pair[i] = (pairs_ant_index, pairs_features, pairs_labels)
pairs_ant_index, pairs_features_raw, pairs_labels = list(zip(*pairs))
pairs_features_raw = np.stack(pairs_features_raw)
pairs_labels = np.squeeze(np.stack(pairs_labels), axis=1)
# Build pair features array (float) from raw features (int)
assert pairs_features_raw[0, :].shape[0] == SIZE_FP_COMPRESSED
pairs_features = np.zeros((len(pairs_ant_index), SIZE_FP))
pairs_features[:, 0:6] = pairs_features_raw[:, 0:6]
pairs_features[:, 6:17] = encode_distance(pairs_features_raw[:, 6])
pairs_features[:, 17:28] = encode_distance(pairs_features_raw[:, 7])
pairs_features[:, 28] = pairs_features_raw[:, 8]
# prepare antecent features
ant_features_raw = np.concatenate([
self.mentions[np.asscalar(idx)][0][np.newaxis, :]
for idx in pairs_ant_index
])
ant_features = np.zeros((pairs_length, SIZE_FS - SIZE_GENRE))
ant_features[:, ant_features_raw[:, 0]] = 1
ant_features[:, 4:15] = encode_distance(ant_features_raw[:, 1])
ant_features[:, 15] = ant_features_raw[:, 2].astype(
float) / ant_features_raw[:, 3].astype(float)
ant_features[:, 16] = ant_features_raw[:, 4]
pairs_features[:, 29:46] = ant_features
# Here we keep the genre
ana_features = np.tile(features, (pairs_length, 1))
pairs_features[:, 46:] = ana_features
ant_spans = np.concatenate([
self.mentions[np.asscalar(idx)][4][np.newaxis, :]
for idx in pairs_ant_index
])
ant_words = np.concatenate([
self.mentions[np.asscalar(idx)][5][np.newaxis, :]
for idx in pairs_ant_index
])
##np.tile（A,B)重复A B次，B可以是列表
ana_spans = np.tile(spans, (pairs_length, 1))
ana_words = np.tile(words, (pairs_length, 1))
##将ndarray转换成tensor
ant_spans = torch.from_numpy(ant_spans).float()
ant_words = torch.from_numpy(ant_words)
ana_spans = torch.from_numpy(ana_spans).float()
ana_words = torch.from_numpy(ana_words)
pairs_features = torch.from_numpy(pairs_features).float()
labels_stack = np.concatenate((pairs_labels, label), axis=0)
assert labels_stack.shape == (pairs_length + 1, )
labels = torch.from_numpy(labels_stack).float()
spans = torch.from_numpy(spans).float()
words = torch.from_numpy(words)
features = torch.from_numpy(features).float()
inputs = (spans, words, features, ant_spans, ant_words, ana_spans,
ana_words, pairs_features)
if self.no_targets:
return inputs
if label == 0:
costs = np.concatenate(
(self.costs['WL'] * (1 - pairs_labels),
[self.costs['FN']])) # Inverse labels: 1=>0, 0=>1
else:
costs = np.concatenate(
(self.costs['FL'] * np.ones_like(pairs_labels), [0]))
assert costs.shape == (pairs_length + 1, )
costs = torch.from_numpy(costs).float()
true_ants_unpad = np.flatnonzero(labels_stack)
if len(true_ants_unpad) == 0:
raise ValueError("Error: no True antecedent for mention")
true_ants = np.pad(true_ants_unpad,
(0, len(pairs_labels) + 1 - len(true_ants_unpad)),
'edge')
assert true_ants.shape == (pairs_length + 1, )
true_ants = torch.from_numpy(true_ants).long()
false_ants_unpad = np.flatnonzero(1 - labels_stack)
assert len(false_ants_unpad) != 0
false_ants = np.pad(false_ants_unpad,
(0, len(pairs_labels) + 1 - len(false_ants_unpad)),
'edge')
assert false_ants.shape == (pairs_length + 1, )
false_ants = torch.from_numpy(false_ants).long()
targets = (labels, costs, true_ants, false_ants)
if debug:
print("Mention", mention_idx)
print("inputs shapes: ", [a.size() for a in inputs])
print("targets shapes: ", [a.size() for a in targets])
return inputs, targets
def __getitem__(self, mention_idx, debug=False):
# add lru cache
"""
Return:
Definitions:
P is the number of antecedent per mention (number of pairs for the mention)
S = 250 is the size of the span vector (averaged word embeddings)
W = 8 is the number of words in a mention (tuned embeddings)
Fp = 70 is the number of features for a pair of mention
Fs = 24 is the number of features of a single mention
if there are some pairs:
inputs = (spans, words, features, ant_spans, ant_words, ana_spans, ana_words, pairs_features)
targets = (labels, costs, true_ants, false_ants)
else:
inputs = (spans, words, features)
targets = (labels, costs, true_ants)
inputs: Tuple of
spans => (S,)
words => (W,)
features => (Fs,)
+ if there are potential antecedents (P > 0):
ant_spans => (P, S) or nothing if no pairs
ant_words => (P, W) or nothing if no pairs
ana_spans => (P, S) or nothing if no pairs
ana_words => (P, W) or nothing if no pairs
pair_features => (P, Fp) or nothing if no pairs
targets: Tuple of
labels => (P+1,)
costs => (P+1,)
true_ant => (P+1,)
+ if there are potential antecedents (P > 0):
false_ant => (P+1,)
"""
#print("PRINTING ALL ARR AT MENTIONS IDX")
#for key, arr in sorted(self.datas.items()) :
# print("<----------------->")
# print(key)
# #print((arr))
# print(type(arr))
# print(arr[mention_idx])
# #print(arr[mention_idx].shape)
# print("<---------------->")
#nangool = input("ALL ARR PRINTED")
#mentions_tuples = [arr[mention_idx,:] for key, arr in sorted(self.datas.items()) if key.startswith("mentions")]
#print("+++++++++++++++++++++++++++++++++++++++++++")
#for arr in mentions_idx_tuples :
# print(arr.shape)
#print("+++++++++++++++++++++++++++++++++++++++++++")
#print(">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>")
#print(mentions_idx_tuples)
#print("<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<")
#mentions_tuples = list(zip(*(arr[mention_idx,:] for key, arr in sorted(self.datas.items()) if key.startswith(u"mentions"))))
#print("`````````````````````````````````````````````````````````")
#for tup in mentions_tuples :
# print(tup)
# print(tup.shape)
# print(tup.shape[0])
#print("``````````````````````````````````````````````````````````")
#mentions_tuples = list(zip(*mentions_idx_tuples))
#print("//////////////////////////////////////////////////////")
#print(mentions_tuples)
#print(mentions_tuples[0])
#print("///////////////////////////////////////////////////////")
#assert [arr.shape[0] for arr in mentions_tuples[0]] == [6, 1, 1, 1, 250, 8] # Cf order of FEATURES_NAMES in conllparser.py
#assert [arr.shape[0] for arr in mentions_tuples] == [6, 1, 1, 1, 250, 8] # Cf order of FEATURES_NAMES in conllparser.py
#features_raw, label, pairs_length, pairs_start_index, spans, words = mentions_tuples#self.mentions[mention_idx]
features_raw, label, pairs_length, pairs_start_index, spans, words = self.mentions[
mention_idx]
pairs_start_index = np.asscalar(pairs_start_index)
pairs_length = np.asscalar(pairs_length)
# Build features array (float) from raw features (int)
assert features_raw.shape[0] == SIZE_FS_COMPRESSED
features = np.zeros((SIZE_FS, ))
features[features_raw[0]] = 1
features[4:15] = encode_distance(features_raw[1])
features[15] = features_raw[2].astype(float) / features_raw[3].astype(
float)
features[16] = features_raw[4]
features[features_raw[5] + 17] = 1
if pairs_length == 0:
spans = torch.from_numpy(spans).float()
words = torch.from_numpy(words)
features = torch.from_numpy(features).float()
inputs = (spans, words, features)
if self.no_targets:
return inputs
true_ant = torch.zeros(1).long() # zeros = indices of true ant
costs = torch.from_numpy((1 - label) * self.costs['FN']).float()
label = torch.from_numpy(label).float()
targets = (label, costs, true_ant)
if debug:
print("inputs shapes: ", [a.size() for a in inputs])
print("targets shapes: ", [a.size() for a in targets])
return inputs, targets
start = pairs_start_index
end = pairs_start_index + pairs_length
#pairs = list(zip(*(arr[start:end] for key, arr in sorted(self.datas.items()) if key.startswith(u"pairs"))))
#assert [arr.shape[0] for arr in pairs[0]] == [1, 9, 1] # Cf order of FEATURES_NAMES in conllparser.py
pairs = self.pairs[start:end]
assert len(pairs) == pairs_length
assert len(
pairs[0]
) == 3 # pair[i] = (pairs_ant_index, pairs_features, pairs_labels)
pairs_ant_index, pairs_features_raw, pairs_labels = list(zip(*pairs))
pairs_features_raw = np.stack(pairs_features_raw)
pairs_labels = np.squeeze(np.stack(pairs_labels), axis=1)
# Build pair features array (float) from raw features (int)
assert pairs_features_raw[0, :].shape[0] == SIZE_FP_COMPRESSED
pairs_features = np.zeros((len(pairs_ant_index), SIZE_FP))
pairs_features[:, 0:6] = pairs_features_raw[:, 0:6]
pairs_features[:, 6:17] = encode_distance(pairs_features_raw[:, 6])
pairs_features[:, 17:28] = encode_distance(pairs_features_raw[:, 7])
pairs_features[:, 28] = pairs_features_raw[:, 8]
# prepare antecent features
ant_features_raw = np.concatenate([
self.mentions[np.asscalar(idx)][0][np.newaxis, :]
for idx in pairs_ant_index
])
#ant_features_raw = np.concatenate([arr[]])
ant_features = np.zeros((pairs_length, SIZE_FS - SIZE_GENRE))
ant_features[:, ant_features_raw[:, 0]] = 1
ant_features[:, 4:15] = encode_distance(ant_features_raw[:, 1])
ant_features[:, 15] = ant_features_raw[:, 2].astype(
float) / ant_features_raw[:, 3].astype(float)
ant_features[:, 16] = ant_features_raw[:, 4]
pairs_features[:, 29:46] = ant_features
# Here we keep the genre
ana_features = np.tile(features, (pairs_length, 1))
pairs_features[:, 46:] = ana_features
ant_spans = np.concatenate([
self.mentions[np.asscalar(idx)][4][np.newaxis, :]
for idx in pairs_ant_index
])
ant_words = np.concatenate([
self.mentions[np.asscalar(idx)][5][np.newaxis, :]
for idx in pairs_ant_index
])
ana_spans = np.tile(spans, (pairs_length, 1))
ana_words = np.tile(words, (pairs_length, 1))
ant_spans = torch.from_numpy(ant_spans).float()
ant_words = torch.from_numpy(ant_words)
ana_spans = torch.from_numpy(ana_spans).float()
ana_words = torch.from_numpy(ana_words)
pairs_features = torch.from_numpy(pairs_features).float()
labels_stack = np.concatenate((pairs_labels, label), axis=0)
assert labels_stack.shape == (pairs_length + 1, )
labels = torch.from_numpy(labels_stack).float()
spans = torch.from_numpy(spans).float()
words = torch.from_numpy(words)
features = torch.from_numpy(features).float()
inputs = (spans, words, features, ant_spans, ant_words, ana_spans,
ana_words, pairs_features)
del spans, words, features, ant_spans, ant_words, ana_spans, ana_words, pairs_features
if self.no_targets:
return inputs
if label == 0:
costs = np.concatenate(
(self.costs['WL'] * (1 - pairs_labels),
[self.costs['FN']])) # Inverse labels: 1=>0, 0=>1
else:
costs = np.concatenate(
(self.costs['FL'] * np.ones_like(pairs_labels), [0]))
assert costs.shape == (pairs_length + 1, )
costs = torch.from_numpy(costs).float()
true_ants_unpad = np.flatnonzero(labels_stack)
if len(true_ants_unpad) == 0:
raise ValueError("Error: no True antecedent for mention")
true_ants = np.pad(true_ants_unpad,
(0, len(pairs_labels) + 1 - len(true_ants_unpad)),
'edge')
assert true_ants.shape == (pairs_length + 1, )
true_ants = torch.from_numpy(true_ants).long()
false_ants_unpad = np.flatnonzero(1 - labels_stack)
assert len(false_ants_unpad) != 0
false_ants = np.pad(false_ants_unpad,
(0, len(pairs_labels) + 1 - len(false_ants_unpad)),
'edge')
assert false_ants.shape == (pairs_length + 1, )
false_ants = torch.from_numpy(false_ants).long()
targets = (labels, costs, true_ants, false_ants)
del labels, costs, true_ants, false_ants
if debug:
print("Mention", mention_idx)
print("inputs shapes: ", [a.size() for a in inputs])
print("targets shapes: ", [a.size() for a in targets])
#lknvf = input("HEY I REACHED THE END OF THIS FUNCTION THAT SHOULDN'T BE HAPPENING")
return inputs, targets
def run_coref_on_mentions(self, mentions):
'''
Run the coreference model on a mentions list
'''
best_ant = {}
best_score = {}
n_ant = 0
#print(mentions)
#inp = np.empty((SIZE_SINGLE_IN, len(mentions)))
#print("SHAPE OF INP SHAPE")
#print(inp.shape)
#yur = input()
#for i, mention_idx in enumerate(mentions):
# mention = self.data[mention_idx]
# print(mention)
# frio = input("mention extraced from data")
# print(type(mention))
# shah = input("mention type")
# print()
# print("mention embedding", mention.embedding.shape)
# inp[:len(mention.embedding), i] = mention.embedding
# inp[:len(mention.embedding), i] = mention.features
# inp[:len(mention.embedding), i] = self.data.genre
mention_idx_list = []
mentions_spans = []
mentions_words = []
mentions_features = []
pairs_ant_idx = []
pairs_features = []
pairs_labels = []
mentions_labels = []
mentions_pairs_start = []
mentions_pairs_length = []
mentions_location = []
mentions_stories = []
n_mentions = 0
total_pairs = 0
#print(mentions)
#oven_sotry = input('MENTIONS PRINTED')
#if debug: print("mentions", self.mentions, str([m.gold_label for m in self.mentions]))
# create 2 for loops, one for single pairs and one for pairs
for mention_idx, antecedents_idx in list(
self.data.get_candidate_pairs(mentions, self.max_dist,
self.max_dist_match)):
n_mentions += 1
doc_id = 1
mention = self.data[mention_idx]
# let's create the story
story_embeds = []
raw_utterances = self.get_utterances()
for utt_index in range(mention.utterance_index):
utt_dealt = raw_utterances[utt_index]
for token in utt_dealt:
# since mention_words_idx works on Mention, we convert every token into a mention
token_word_idx = word_idx_finder(self.embed_extractor,
token.text)
#story_embeds.append(token_embed.tolist())
story_embeds.append(token_word_idx)
final_utt_dealt = raw_utterances[mention.utterance_index]
for token_index in range(mention.start):
token_word_idx = word_idx_finder(
self.embed_extractor, final_utt_dealt[token_index].text)
#story_embeds.append(token_embed.tolist())
story_embeds.append(token_word_idx)
mentions_stories.append(story_embeds)
mention_idx_list.append(mention_idx)
mentions_spans.append(mention.spans_embeddings)
w_idx = mention_words_idx(self.embed_extractor, mention)
if w_idx is None:
print("error in", self.name, self.part,
mention.utterance_index)
mentions_words.append(w_idx)
mentions_features.append(
self.get_single_mention_features_conll(mention))
mentions_location.append([
mention.start, mention.end, mention.utterance_index,
mention_idx, doc_id
])
ants = [self.data.mentions[ant_idx] for ant_idx in antecedents_idx]
# Some display functions
#tuy = input()
#print("************************************************************************************************")
#print("MENTION IDX,",mention_idx)
#print("MENTION REFRED,",mention)
#print("MENTION SPANS,",mention.spans_embeddings.shape)
#print("MENTION FEATURES,",self.get_single_mention_features_conll(mention))
#print("MENTION LOCATION,",[mention.start,mention.end,mention.utterance_index,mention_idx,doc_id])
#print("ANTS ,",ants)
#print("*************************************************************************************************")
#hua = input()
no_antecedent = not any(
ant.gold_label == mention.gold_label
for ant in ants) or mention.gold_label is None
if antecedents_idx:
pairs_ant_idx += [idx for idx in antecedents_idx]
pairs_features += [
self.get_pair_mentions_features_conll(ant, mention)
for ant in ants
]
ant_labels = [0 for ant in ants] if no_antecedent else [
1 if ant.gold_label == mention.gold_label else 0
for ant in ants
]
pairs_labels += ant_labels
mentions_labels.append(1 if no_antecedent else 0)
mentions_pairs_start.append(total_pairs)
total_pairs += len(ants)
mentions_pairs_length.append(len(ants))
out_dict = {
FEATURES_NAMES[0]:
mentions_features,
FEATURES_NAMES[1]:
mentions_labels,
FEATURES_NAMES[2]:
mentions_pairs_length,
FEATURES_NAMES[3]:
mentions_pairs_start,
FEATURES_NAMES[4]:
mentions_spans,
FEATURES_NAMES[5]:
mentions_words,
#FEATURES_NAMES[6]: pairs_ant_idx if pairs_ant_idx else None,
FEATURES_NAMES[6]:
pairs_ant_idx if pairs_ant_idx else list(),
#FEATURES_NAMES[7]: pairs_features if pairs_features else None,
FEATURES_NAMES[7]:
pairs_features if pairs_features else list(),
#FEATURES_NAMES[8]: pairs_labels if pairs_labels else None,
FEATURES_NAMES[8]:
pairs_labels if pairs_labels else list(),
FEATURES_NAMES[9]:
mentions_stories
}
gathering_dict = dict((feat, None) for feat in FEATURES_NAMES)
n_mentions_list = []
pairs_ant_index = 0
pairs_start_index = 0
for n, p, arrays_dict in tqdm([(n_mentions, total_pairs, out_dict)]):
#print(out_dict)
#pizza_hut = input('OUT DICT PRINTED')
#print(arrays_dict)
#dominoes = input('ARRAYS DICT PRINTED')
for f in FEATURES_NAMES:
if gathering_dict[f] is None:
gathering_dict[f] = arrays_dict[f]
else:
if f == FEATURES_NAMES[6]:
array = [a + pairs_ant_index for a in arrays_dict[f]]
elif f == FEATURES_NAMES[3]:
array = [a + pairs_start_index for a in arrays_dict[f]]
else:
array = arrays_dict[f]
gathering_dict[f] += array
pairs_ant_index += n
pairs_start_index += p
n_mentions_list.append(n)
mention_feature_dict = dict()
pairs_feature_dict = dict()
train_phase = True
for feature in FEATURES_NAMES[:10]:
print("Building numpy array for", feature, "length",
len(gathering_dict[feature]))
if feature != "mentions_spans":
#array = np.array(gathering_dict[feature])
# check if we are dealing with length of memories
if feature == "mentions_stories" or feature == "pairs_stories":
gathering_array = []
max_story_len = 200
for story in gathering_dict[feature]:
#print(len(story[0]))
#print(len(story[1]))
#random_pause = input()
if len(story) > 200:
final_story = story[-200:]
else:
number_to_append = max(0,
max_story_len - len(story))
#number_to_append = min(number_to_append,50)
final_story = story + number_to_append * [0]
#print(final_story)
#print(len(final_story))
#random_pause = input()
gathering_array.append(final_story)
array = np.array(gathering_array)
print(array.shape)
else:
array = np.array(gathering_dict[feature])
if array.ndim == 1:
print("expand_dims for feature, ", feature)
array = np.expand_dims(array, axis=1)
else:
array = np.stack(gathering_dict[feature])
# check_numpy_array(feature, array, n_mentions_list)
print("Saving numpy", feature, "size", array.shape)
#array_save = input()
if feature.startswith("mentions"):
mention_feature_dict[feature] = array
if feature.startswith("pairs"):
pairs_feature_dict[feature] = array
# zip it with pairs dict
self.mentions = list(
zip(*(arr for key, arr in sorted(mention_feature_dict.items()))))
self.pairs = list(
zip(*(arr for key, arr in sorted(pairs_feature_dict.items()))))
#print("LEN OF PAIRS IS,",len(self.pairs))
#jsk = input("PRINTING THE PAIRS")
#print(self.pairs)
#sdghr = input("ALL PAIRS PRINTED")
#print("MENTION PAIRS LENGTH IS,",mention_feature_dict['mentions_pairs_length'])
#victoria = input()
for i in range(len(mention_feature_dict[FEATURES_NAMES[0]])):
mention_idx = mention_idx_list[i]
features_raw = mention_feature_dict['mentions_features'][i, :]
#print("FEATUERES_RAW_PRINTED_is,",features_raw)
label = mention_feature_dict['mentions_labels'][i, :]
pairs_length = mention_feature_dict['mentions_pairs_length'][i, :]
pairs_start_index = mention_feature_dict[
'mentions_pairs_start_index'][i]
mentions_stories = mention_feature_dict['mentions_stories'][i]
spans = mention_feature_dict['mentions_spans'][i, :]
words = mention_feature_dict['mentions_words'][i, :]
pairs_start_index = np.asscalar(pairs_start_index)
pairs_length = np.asscalar(pairs_length)
# Build features array (float) from raw features (int)
assert features_raw.shape[0] == SIZE_FS_COMPRESSED
features = np.zeros((SIZE_FS, ))
features[features_raw[0]] = 1
features[4:15] = encode_distance(features_raw[1])
features[15] = features_raw[2].astype(
float) / features_raw[3].astype(float)
features[16] = features_raw[4]
features[features_raw[5] + 17] = 1
#print("====================================<>============================================")
#print("TYPE OF SPANS,",type(spans))
#print("TYPE OF WORDS,",type(words))
#print("TYPE OF FEATURES,",type(features))
#print("====================================<>============================================")
spans = spans[np.newaxis, :]
print("PRINTING SHAPE OF WORDS")
print(words.shape)
words = words[np.newaxis, :]
features = features[np.newaxis, :]
mentions_stories = mentions_stories[np.newaxis, :]
spans = torch.from_numpy(spans).float()
words = torch.from_numpy(words)
features = torch.from_numpy(features).float()
mentions_stories = torch.from_numpy(mentions_stories)
#print(mentions_stories.size())
#print(words.size())
#kake = input("size of mentions stories is ")
# inputs for the single mentions
#print("SINGLE SCORES COMPUTING")
single_inputs = (spans, words, features, mentions_stories)
score = self.coref_model.get_multiple_single_score(
single_inputs).tolist()[0][0]
#print("PRINTING SINGLE SCORE")
#print(score)
#sgbet = input("SINGLE SCORE PRINTED")
self.mentions_single_scores[mention_idx] = score
best_score[mention_idx] = score - 50 * (self.greedyness - 0.5)
#print("SINGLE SCORES COMPUTED")
if pairs_length == 0:
continue
start = pairs_start_index
end = pairs_start_index + pairs_length
pairs = self.pairs[start:end]
#print("START IS,",start)
#print("END IS,",end)
#print("PAIRS LENGTH,",pairs_length)
#print("LEN OF PAIRS IS,",len(pairs))
assert len(pairs) == pairs_length
assert len(
pairs[0]
) == 3 # pair[i] = (pairs_ant_index, pairs_features, pairs_labels)
pairs_ant_index, pairs_features_raw, pairs_labels = list(
zip(*pairs))
pairs_features_raw = np.stack(pairs_features_raw)
pairs_labels = np.squeeze(np.stack(pairs_labels), axis=1)
# Build pair features array (float) from raw features (int)
assert pairs_features_raw[0, :].shape[0] == SIZE_FP_COMPRESSED
pairs_features = np.zeros((len(pairs_ant_index), SIZE_FP))
pairs_features[:, 0:6] = pairs_features_raw[:, 0:6]
pairs_features[:, 6:17] = encode_distance(pairs_features_raw[:, 6])
pairs_features[:, 17:28] = encode_distance(pairs_features_raw[:,
7])
pairs_features[:, 28] = pairs_features_raw[:, 8]
# prepare antecent features
# printing antecedent features
#hsya = input("PRINTING DATA MENTIONS")
#print(self.data.mentions)
#uba = input("PRINTED DATA MENTIONS")
ant_features_raw = np.concatenate([
self.mentions[np.asscalar(idx)][0][np.newaxis, :]
for idx in pairs_ant_index
])
ant_features = np.zeros((pairs_length, SIZE_FS - SIZE_GENRE))
ant_features[:, ant_features_raw[:, 0]] = 1
ant_features[:, 4:15] = encode_distance(ant_features_raw[:, 1])
ant_features[:, 15] = ant_features_raw[:, 2].astype(
float) / ant_features_raw[:, 3].astype(float)
ant_features[:, 16] = ant_features_raw[:, 4]
pairs_features[:, 29:46] = ant_features
# Here we keep the genre
ana_features = np.tile(features, (pairs_length, 1))
pairs_features[:, 46:] = ana_features
ant_spans = np.concatenate([
self.mentions[np.asscalar(idx)][4][np.newaxis, :]
for idx in pairs_ant_index
])
ant_words = np.concatenate([
self.mentions[np.asscalar(idx)][6][np.newaxis, :]
for idx in pairs_ant_index
])
ana_spans = np.tile(spans, (pairs_length, 1))
ana_words = np.tile(words, (pairs_length, 1))
ant_spans = ant_spans[np.newaxis, :]
ant_words = ant_words[np.newaxis, :]
ana_spans = ana_spans[np.newaxis, :]
ana_words = ana_words[np.newaxis, :]
pairs_features = pairs_features[np.newaxis, :]
ant_spans = torch.from_numpy(ant_spans).float()
ant_words = torch.from_numpy(ant_words)
ana_spans = torch.from_numpy(ana_spans).float()
ana_words = torch.from_numpy(ana_words)
pairs_features = torch.from_numpy(pairs_features).float()
labels_stack = np.concatenate((pairs_labels, label), axis=0)
assert labels_stack.shape == (pairs_length + 1, )
labels = torch.from_numpy(labels_stack).float()
# inputs for the pairs of mentions
pairs_inputs = (spans, words, features, ant_spans, ant_words,
ana_spans, ana_words, pairs_features,
mentions_stories)
#print("PAIRS INPUT CREATED")
score = self.coref_model.get_multiple_pair_score(
pairs_inputs).tolist()[0][:-1]
#print("SCORES GOT IS ")
#print(score)
#hutys = input("SCORES PRINTED")
for ant_idx, s in zip(pairs_ant_idx, score):
self.mentions_pairs_scores[mention_idx][ant_idx] = s
if s > best_score[mention_idx]:
best_score[mention_idx] = s
best_ant[mention_idx] = ant_idx
if mention_idx in best_ant:
n_ant += 1
self._merge_coreference_clusters(best_ant[mention_idx],
mention_idx)
#score = self.coref_model.get_multiple_single_score(inp.T)
#score = self.coref_model(tuple(mentions_spans,mentions_words,mentions_features
# find a way to access the mention_idx
return (n_ant, best_ant)