Rooba
/
spaCy
mirror of https://github.com/explosion/spaCy.git
1
1
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Merge from develop

This commit is contained in:
Matthw Honnibal 2020-05-20 12:27:31 +02:00
parent 7f5715a081
commit 24efd54a42
3 changed files with 12 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
spacy/cli/train_from_config.py
View File

@ -224,8 +224,9 @@ def train_from_config(
def create_train_batches(nlp, corpus, cfg): def create_train_batches(nlp, corpus, cfg):
is_first = True
while True: while True:
train_examples = list(corpus.train_dataset( train_examples = corpus.train_dataset(
nlp, nlp,
noise_level=0.0, noise_level=0.0,
orth_variant_level=cfg["orth_variant_level"], orth_variant_level=cfg["orth_variant_level"],
@ -323,6 +324,8 @@ def train_while_improving(
for subbatch in subdivide_batch(batch, accumulate_gradient): for subbatch in subdivide_batch(batch, accumulate_gradient):
nlp.update(subbatch, drop=dropout, losses=losses, sgd=False) nlp.update(subbatch, drop=dropout, losses=losses, sgd=False)
for name, proc in nlp.pipeline: for name, proc in nlp.pipeline:
for name, proc in nlp.pipeline:
if hasattr(proc, "model"):
proc.model.finish_update(optimizer) proc.model.finish_update(optimizer)
optimizer.step_schedules() optimizer.step_schedules()
if not (step % eval_frequency): if not (step % eval_frequency):

6
spacy/syntax/_parser_model.pyx
View File

@ -474,7 +474,11 @@ cdef class precompute_hiddens:
# This will usually be on GPU # This will usually be on GPU
d_best = ops.asarray(d_best) d_best = ops.asarray(d_best)
# Fix nans (which can occur from unseen classes.) # Fix nans (which can occur from unseen classes.)
d_best[ops.xp.isnan(d_best)] = 0. try:
d_best[ops.xp.isnan(d_best)] = 0.
except:
print(ops.xp.isnan(d_best))
raise
if self.activation == "maxout": if self.activation == "maxout":
mask_ = ops.asarray(mask) mask_ = ops.asarray(mask)
return ops.backprop_maxout(d_best, mask_, self.nP) return ops.backprop_maxout(d_best, mask_, self.nP)

13
spacy/util.py
View File

@ -598,16 +598,9 @@ def minibatch_by_words(examples, size, tuples=True, count_words=len, tolerance=0
try: try:
example = next(examples) example = next(examples)
except StopIteration: except StopIteration:
if oversize: if batch:
examples = iter(oversize) yield batch
oversize = [] return
if batch:
yield batch
break
else:
if batch:
yield batch
return
n_words = count_words(example.doc) n_words = count_words(example.doc)
if n_words < (batch_size + tol_size): if n_words < (batch_size + tol_size):
batch_size -= n_words batch_size -= n_words