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

Optimizer defaults (#5244) 

* set optimizer defaults to mimic thinc 7 + bump to dev6

* larger error range for senter overfitting test
This commit is contained in:
Sofie Van Landeghem 2020-04-03 13:02:46 +02:00 committed by GitHub
parent b71a11ff6d
commit b2e93be867
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 5 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
spacy/about.py
View File

@ -1,6 +1,6 @@
# fmt: off # fmt: off
__title__ = "spacy" __title__ = "spacy"
__version__ = "3.0.0.dev4" __version__ = "3.0.0.dev6"
__release__ = True __release__ = True
__download_url__ = "https://github.com/explosion/spacy-models/releases/download" __download_url__ = "https://github.com/explosion/spacy-models/releases/download"
__compatibility__ = "https://raw.githubusercontent.com/explosion/spacy-models/master/compatibility.json" __compatibility__ = "https://raw.githubusercontent.com/explosion/spacy-models/master/compatibility.json"

2
spacy/tests/pipeline/test_senter.py
View File

@ -33,7 +33,7 @@ def test_overfitting_IO():
for i in range(200): for i in range(200):
losses = {} losses = {}
nlp.update(TRAIN_DATA, sgd=optimizer, losses=losses) nlp.update(TRAIN_DATA, sgd=optimizer, losses=losses)
assert losses["senter"] < 0.0001 assert losses["senter"] < 0.001
# test the trained model # test the trained model
test_text = "I like purple eggs. They eat ham. You like yellow eggs." test_text = "I like purple eggs. They eat ham. You like yellow eggs."

4
spacy/util.py
View File

@ -790,7 +790,8 @@ def create_default_optimizer():
beta2 = env_opt("optimizer_B2", 0.999) beta2 = env_opt("optimizer_B2", 0.999)
eps = env_opt("optimizer_eps", 1e-8) eps = env_opt("optimizer_eps", 1e-8)
L2 = env_opt("L2_penalty", 1e-6) L2 = env_opt("L2_penalty", 1e-6)
grad_clip = env_opt("grad_norm_clip", 1.0) grad_clip = env_opt("grad_norm_clip", 10.0)
L2_is_weight_decay = env_opt("L2_is_weight_decay", False)
optimizer = Adam( optimizer = Adam(
learn_rate, learn_rate,
L2=L2, L2=L2,
@ -799,5 +800,6 @@ def create_default_optimizer():
eps=eps, eps=eps,
ops=ops, ops=ops,
grad_clip=grad_clip, grad_clip=grad_clip,
L2_is_weight_decay=L2_is_weight_decay,
) )
return optimizer return optimizer