From 2abb4ba9db0d0ec074a7336be8a7395da78eaaa4 Mon Sep 17 00:00:00 2001
From: Matthew Honnibal <honnibal+gh@gmail.com>
Date: Thu, 24 Sep 2020 18:13:39 +0200
Subject: [PATCH] Make a pre-check to speed up alignment cache (#6139)
* Dirty trick to fast-track alignment cache
* Improve alignment cache check
* Fix header
* Fix align cache
* Fix align logic
---
spacy/training/example.pxd | 3 +++
spacy/training/example.pyx | 36 +++++++++++++++++++++++++++++-------
2 files changed, 32 insertions(+), 7 deletions(-)
diff --git a/spacy/training/example.pxd b/spacy/training/example.pxd
index e06e36287..49e239757 100644
--- a/spacy/training/example.pxd
+++ b/spacy/training/example.pxd
@@ -1,4 +1,5 @@
from ..tokens.doc cimport Doc
+from libc.stdint cimport uint64_t
cdef class Example:
@@ -7,3 +8,5 @@ cdef class Example:
cdef readonly object _cached_alignment
cdef readonly object _cached_words_x
cdef readonly object _cached_words_y
+ cdef readonly uint64_t _x_sig
+ cdef readonly uint64_t _y_sig
diff --git a/spacy/training/example.pyx b/spacy/training/example.pyx
index 1e7bea5df..6a9815c44 100644
--- a/spacy/training/example.pyx
+++ b/spacy/training/example.pyx
@@ -1,6 +1,7 @@
from collections import Iterable as IterableInstance
import warnings
import numpy
+from murmurhash.mrmr cimport hash64
from ..tokens.doc cimport Doc
from ..tokens.span cimport Span
@@ -97,15 +98,36 @@ cdef class Example:
@property
def alignment(self):
- words_x = [token.text for token in self.x]
- words_y = [token.text for token in self.y]
- if self._cached_alignment is None or \
- words_x != self._cached_words_x or \
- words_y != self._cached_words_y:
- self._cached_alignment = Alignment.from_strings(words_x, words_y)
+ x_sig = hash64(self.x.c, sizeof(self.x.c[0]) * self.x.length, 0)
+ y_sig = hash64(self.y.c, sizeof(self.y.c[0]) * self.y.length, 0)
+ if self._cached_alignment is None:
+ words_x = [token.text for token in self.x]
+ words_y = [token.text for token in self.y]
+ self._x_sig = x_sig
+ self._y_sig = y_sig
self._cached_words_x = words_x
self._cached_words_y = words_y
- return self._cached_alignment
+ self._cached_alignment = Alignment.from_strings(words_x, words_y)
+ return self._cached_alignment
+ elif self._x_sig == x_sig and self._y_sig == y_sig:
+ # If we have a cached alignment, check whether the cache is invalid
+ # due to retokenization. To make this check fast in loops, we first
+ # check a hash of the TokenC arrays.
+ return self._cached_alignment
+ else:
+ words_x = [token.text for token in self.x]
+ words_y = [token.text for token in self.y]
+ if words_x == self._cached_words_x and words_y == self._cached_words_y:
+ self._x_sig = x_sig
+ self._y_sig = y_sig
+ return self._cached_alignment
+ else:
+ self._cached_alignment = Alignment.from_strings(words_x, words_y)
+ self._cached_words_x = words_x
+ self._cached_words_y = words_y
+ self._x_sig = x_sig
+ self._y_sig = y_sig
+ return self._cached_alignment
def get_aligned(self, field, as_string=False):
"""Return an aligned array for a token attribute."""