mirror of https://github.com/explosion/spaCy.git
Dont randomize cuts in parser
This commit is contained in:
parent
709fc5e4ad
commit
f25761e513
|
@ -280,11 +280,12 @@ cdef class Parser:
|
||||||
[eg.predicted for eg in examples])
|
[eg.predicted for eg in examples])
|
||||||
if self.cfg["update_with_oracle_cut_size"] >= 1:
|
if self.cfg["update_with_oracle_cut_size"] >= 1:
|
||||||
# Chop sequences into lengths of this many transitions, to make the
|
# Chop sequences into lengths of this many transitions, to make the
|
||||||
# batch uniform length. We randomize this to overfit less.
|
# batch uniform length.
|
||||||
|
# We used to randomize this, but it's not clear that actually helps?
|
||||||
cut_size = self.cfg["update_with_oracle_cut_size"]
|
cut_size = self.cfg["update_with_oracle_cut_size"]
|
||||||
states, golds, max_steps = self._init_gold_batch(
|
states, golds, max_steps = self._init_gold_batch(
|
||||||
examples,
|
examples,
|
||||||
max_length=numpy.random.choice(range(5, cut_size))
|
max_length=cut_size
|
||||||
)
|
)
|
||||||
else:
|
else:
|
||||||
states, golds, _ = self.moves.init_gold_batch(examples)
|
states, golds, _ = self.moves.init_gold_batch(examples)
|
||||||
|
|
Loading…
Reference in New Issue