explosion/spaCy
1
1
Fork 0
mirror of https://github.com/explosion/spaCy.git synced 2024-12-26 18:06:29 +03:00
Code Issues Packages Projects Releases Wiki Activity

speeding up training

Browse Source
This commit is contained in:
svlandeg 2019-06-12 13:37:05 +02:00
parent 66813a1fdc
commit 6521cfa132
2 changed files with 57 additions and 43 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

25
examples/pipeline/wiki_entity_linking/wiki_nel_pipeline.py
View File

@ -115,6 +115,7 @@ def run_pipeline():
# STEP 6: create the entity linking pipe # STEP 6: create the entity linking pipe
if train_pipe: if train_pipe:
print("STEP 6: training Entity Linking pipe", datetime.datetime.now())
train_limit = 100 train_limit = 100
dev_limit = 20 dev_limit = 20
print("Training on", train_limit, "articles") print("Training on", train_limit, "articles")
@ -147,6 +148,7 @@ def run_pipeline():
with nlp.disable_pipes(*other_pipes): with nlp.disable_pipes(*other_pipes):
for batch in batches: for batch in batches:
try:
docs, golds = zip(*batch) docs, golds = zip(*batch)
nlp.update( nlp.update(
docs, docs,
@ -154,6 +156,16 @@ def run_pipeline():
drop=DROPOUT, drop=DROPOUT,
losses=losses, losses=losses,
) )
except Exception as e:
print("Error updating batch", e)
print("Epoch, train loss", itn, round(losses['entity_linker'], 2))
# baseline using only prior probabilities
el_pipe.context_weight = 0
el_pipe.prior_weight = 1
dev_acc_0_1 = _measure_accuracy(dev_data, el_pipe)
train_acc_0_1 = _measure_accuracy(train_data, el_pipe)
# print(" measuring accuracy 1-1") # print(" measuring accuracy 1-1")
el_pipe.context_weight = 1 el_pipe.context_weight = 1
@ -161,19 +173,13 @@ def run_pipeline():
dev_acc_1_1 = _measure_accuracy(dev_data, el_pipe) dev_acc_1_1 = _measure_accuracy(dev_data, el_pipe)
train_acc_1_1 = _measure_accuracy(train_data, el_pipe) train_acc_1_1 = _measure_accuracy(train_data, el_pipe)
# print(" measuring accuracy 0-1")
el_pipe.context_weight = 0
el_pipe.prior_weight = 1
dev_acc_0_1 = _measure_accuracy(dev_data, el_pipe)
train_acc_0_1 = _measure_accuracy(train_data, el_pipe)
# print(" measuring accuracy 1-0") # print(" measuring accuracy 1-0")
el_pipe.context_weight = 1 el_pipe.context_weight = 1
el_pipe.prior_weight = 0 el_pipe.prior_weight = 0
dev_acc_1_0 = _measure_accuracy(dev_data, el_pipe) dev_acc_1_0 = _measure_accuracy(dev_data, el_pipe)
train_acc_1_0 = _measure_accuracy(train_data, el_pipe) train_acc_1_0 = _measure_accuracy(train_data, el_pipe)
print("Epoch, train loss, train/dev acc, 1-1, 0-1, 1-0:", itn, round(losses['entity_linker'], 2), print("train/dev acc, 1-1, 0-1, 1-0:" ,
round(train_acc_1_1, 2), round(train_acc_0_1, 2), round(train_acc_1_0, 2), "/", round(train_acc_1_1, 2), round(train_acc_0_1, 2), round(train_acc_1_0, 2), "/",
round(dev_acc_1_1, 2), round(dev_acc_0_1, 2), round(dev_acc_1_0, 2)) round(dev_acc_1_1, 2), round(dev_acc_0_1, 2), round(dev_acc_1_0, 2))
@ -193,10 +199,10 @@ def _measure_accuracy(data, el_pipe):
docs = [d for d, g in data] docs = [d for d, g in data]
docs = el_pipe.pipe(docs) docs = el_pipe.pipe(docs)
golds = [g for d, g in data] golds = [g for d, g in data]
for doc, gold in zip(docs, golds): for doc, gold in zip(docs, golds):
try:
correct_entries_per_article = dict() correct_entries_per_article = dict()
for entity in gold.links: for entity in gold.links:
start, end, gold_kb = entity start, end, gold_kb = entity
@ -214,6 +220,9 @@ def _measure_accuracy(data, el_pipe):
else: else:
incorrect += 1 incorrect += 1
except Exception as e:
print("Error assessing accuracy", e)
if correct == incorrect == 0: if correct == incorrect == 0:
return 0 return 0

5
spacy/pipeline/pipes.pyx
View File

@ -1220,8 +1220,13 @@ class EntityLinker(Pipe):
def predict(self, docs): def predict(self, docs):
self.require_model() self.require_model()
if isinstance(docs, Doc):
docs = [docs]
final_entities = list() final_entities = list()
final_kb_ids = list() final_kb_ids = list()
for i, article_doc in enumerate(docs): for i, article_doc in enumerate(docs):
doc_encoding = self.article_encoder([article_doc]) doc_encoding = self.article_encoder([article_doc])
for ent in article_doc.ents: for ent in article_doc.ents: