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

Fix gradient in fine_tune

This commit is contained in:
Matthew Honnibal 2017-08-14 14:55:47 -05:00
parent 0ae045256d
commit 335fa8b05c
1 changed files with 3 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
spacy/_ml.py
View File

@ -377,16 +377,16 @@ def fine_tune(embedding, combine=None):
lengths) lengths)
def fine_tune_bwd(d_output, sgd=None): def fine_tune_bwd(d_output, sgd=None):
bp_vecs(d_output, sgd=sgd) bp_vecs([d_o * model.d_mix[0] for d_o in d_output], sgd=sgd)
flat_grad = model.ops.flatten(d_output) flat_grad = model.ops.flatten(d_output)
model.d_mix[1] += flat_tokvecs.dot(flat_grad.T).sum() model.d_mix[1] += flat_tokvecs.dot(flat_grad.T).sum()
model.d_mix[0] += flat_vecs.dot(flat_grad.T).sum() model.d_mix[0] += flat_vecs.dot(flat_grad.T).sum()
sgd(model._mem.weights, model._mem.gradient, key=model.id) sgd(model._mem.weights, model._mem.gradient, key=model.id)
return d_output return [d_o * model.d_mix[1] for d_o in d_output]
return output, fine_tune_bwd return output, fine_tune_bwd
model = wrap(fine_tune_fwd, embedding) model = wrap(fine_tune_fwd, embedding)
model.mix = model._mem.add((model.id, 'mix'), (2,)) model.mix = model._mem.add((model.id, 'mix'), (2,))
model.mix.fill(1.) model.mix.fill(0.5)
model.d_mix = model._mem.add_gradient((model.id, 'd_mix'), (model.id, 'mix')) model.d_mix = model._mem.add_gradient((model.id, 'd_mix'), (model.id, 'mix'))
return model return model