Merge branch 'develop' of https://github.com/explosion/spaCy into develop

This commit is contained in:
Matthew Honnibal 2017-08-20 12:55:39 -05:00
commit 7a6edeea68
2 changed files with 13 additions and 12 deletions

View File

@ -359,8 +359,6 @@ def get_token_vectors(tokens_attrs_vectors, drop=0.):
def backward(d_output, sgd=None):
return (tokens, d_output)
return vectors, backward
def fine_tune(embedding, combine=None):
if combine is not None:
raise NotImplementedError(
@ -372,22 +370,25 @@ def fine_tune(embedding, combine=None):
vecs, bp_vecs = embedding.begin_update(docs, drop=drop)
flat_tokvecs = embedding.ops.flatten(tokvecs)
flat_vecs = embedding.ops.flatten(vecs)
alpha = model.mix
minus = 1-model.mix
output = embedding.ops.unflatten(
(model.mix[0] * flat_vecs + model.mix[1] * flat_tokvecs),
lengths)
(alpha * flat_tokvecs + minus * flat_vecs), lengths)
def fine_tune_bwd(d_output, sgd=None):
bp_vecs(d_output, sgd=sgd)
flat_grad = model.ops.flatten(d_output)
model.d_mix[1] += flat_tokvecs.dot(flat_grad.T).sum()
model.d_mix[0] += flat_vecs.dot(flat_grad.T).sum()
if sgd is not None:
model.d_mix += flat_tokvecs.dot(flat_grad.T).sum()
model.d_mix += 1-flat_vecs.dot(flat_grad.T).sum()
bp_vecs([d_o * minus for d_o in d_output], sgd=sgd)
d_output = [d_o * alpha for d_o in d_output]
sgd(model._mem.weights, model._mem.gradient, key=model.id)
model.mix = model.ops.xp.minimum(model.mix, 1.0)
return d_output
return output, fine_tune_bwd
model = wrap(fine_tune_fwd, embedding)
model.mix = model._mem.add((model.id, 'mix'), (2,))
model.mix.fill(1.)
model.mix = model._mem.add((model.id, 'mix'), (1,))
model.mix.fill(0.0)
model.d_mix = model._mem.add_gradient((model.id, 'd_mix'), (model.id, 'mix'))
return model

View File

@ -94,7 +94,7 @@ def train(cmd, lang, output_dir, train_data, dev_data, n_iter=20, n_sents=0,
docs, golds = zip(*batch)
nlp.update(docs, golds, sgd=optimizer,
drop=next(dropout_rates), losses=losses,
update_tensors=True)
update_shared=True)
pbar.update(sum(len(doc) for doc in docs))
with nlp.use_params(optimizer.averages):