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

Use ftrl training, to learn compressed model.

This commit is contained in:
Matthew Honnibal 2017-03-09 18:43:21 -06:00
parent f71eeef9bb
commit c62da02344
1 changed files with 15 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
spacy/syntax/parser.pyx
View File

@ -124,7 +124,7 @@ cdef class Parser:
elif 'features' not in cfg: elif 'features' not in cfg:
cfg['features'] = self.feature_templates cfg['features'] = self.feature_templates
self.model = ParserModel(cfg['features']) self.model = ParserModel(cfg['features'])
self.model.l1_penalty = 1e-7 self.model.l1_penalty = cfg.get('L1', 0.0)
self.cfg = cfg self.cfg = cfg
@ -263,10 +263,10 @@ cdef class Parser:
self.model.time += 1 self.model.time += 1
guess = VecVec.arg_max_if_true(eg.c.scores, eg.c.is_valid, eg.c.nr_class) guess = VecVec.arg_max_if_true(eg.c.scores, eg.c.is_valid, eg.c.nr_class)
if eg.c.costs[guess] > 0: if eg.c.costs[guess] > 0:
best = VecVec.arg_max_if_zero(eg.c.scores, eg.c.costs, eg.c.nr_class) best = arg_max_if_gold(eg.c.scores, eg.c.costs, eg.c.nr_class)
for feat in eg.c.features[:eg.c.nr_feat]: for feat in eg.c.features[:eg.c.nr_feat]:
self.model.update_weight_ftrl(feat.key, best, -feat.value * eg.costs[guess]) self.model.update_weight_ftrl(feat.key, best, -feat.value * eg.c.costs[guess])
self.model.update_weight_ftrl(feat.key, guess, feat.value * eg.costs[guess]) self.model.update_weight_ftrl(feat.key, guess, feat.value * eg.c.costs[guess])
action = self.moves.c[guess] action = self.moves.c[guess]
action.do(stcls.c, action.label) action.do(stcls.c, action.label)
@ -392,6 +392,14 @@ class ParserStateError(ValueError):
"Please include the text that the parser failed on, which is:\n" "Please include the text that the parser failed on, which is:\n"
"%s" % repr(doc.text)) "%s" % repr(doc.text))
cdef int arg_max_if_gold(const weight_t* scores, const weight_t* costs, int n) nogil:
cdef int best = -1
for i in range(n):
if costs[i] <= 0:
if best == -1 or scores[i] > scores[best]:
best = i
return best
cdef int _arg_max_clas(const weight_t* scores, int move, const Transition* actions, cdef int _arg_max_clas(const weight_t* scores, int move, const Transition* actions,
int nr_class) except -1: int nr_class) except -1: