Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fix indexing error in word2vec_inner.pyx #3136

Merged
merged 7 commits into from
May 13, 2021
Merged

Fix indexing error in word2vec_inner.pyx #3136

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
c6b47ba
Add explicit typecasting for the W2V cython code
May 9, 2021
83f1ae4
Add explicit typecasting for the W2V cython code
May 9, 2021
ee4b103
Unifying the types
May 10, 2021
e6fea03
Fix minor code style issues
May 10, 2021
a443551
Code refactoring
May 10, 2021
f134e36
Update CHANGELOG.md
mpenkov May 13, 2021
201979f
Merge branch 'develop' into fix-largevector
mpenkov May 13, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Empty file added 16
View file
Open in desktop
Empty file.
1 change: 1 addition & 0 deletions CHANGELOG.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,6 +8,7 @@ Changes
- Improve & unify docs for dirichlet priors (PR [#3125](https://github.com/RaRe-Technologies/gensim/pull/3125), [@jonaschn](https://github.com/jonaschn))
- Materialize and copy the corpus passed to SoftCosineSimilarity (PR [#3128](https://github.com/RaRe-Technologies/gensim/pull/3128), [@Witiko](https://github.com/Witiko))
- [#3115](https://github.com/RaRe-Technologies/gensim/pull/3115): Make LSI dispatcher CLI param for number of jobs optional, by [@robguinness](https://github.com/robguinness))
- [#3136](https://github.com/RaRe-Technologies/gensim/pull/3136): fix indexing error in word2vec_inner.pyx, by [@bluekura](https://github.com/bluekura)
- Update link to Hoffman paper (online VB LDA) (PR [#3133](https://github.com/RaRe-Technologies/gensim/pull/3133), [@jonaschn](https://github.com/jonaschn))
- fix bug when loading saved Phrases model (PR [#3116](https://github.com/RaRe-Technologies/gensim/pull/3116), [@aloknayak29](https://github.com/aloknayak29))
- Optimize performance of Author-Topic model (PR [#2978](https://github.com/RaRe-Technologies/gensim/pull/2978), [@horpto](https://github.com/horpto))
Expand Down
32 changes: 18 additions & 14 deletions gensim/models/word2vec_inner.pyx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -71,6 +71,10 @@ cdef void our_saxpy_noblas(const int *N, const float *alpha, const float *X, con
for i from 0 <= i < N[0] by 1:
Y[i * (incY[0])] = (alpha[0]) * X[i * (incX[0])] + Y[i * (incY[0])]

cdef long long _mul(const np.uint32_t a, const int b) nogil:
"""Safe multiplication of ints with explict typecasting"""
return <long long>a * <long long>b

cdef void w2v_fast_sentence_sg_hs(
const np.uint32_t *word_point, const np.uint8_t *word_code, const int codelen,
REAL_t *syn0, REAL_t *syn1, const int size,
Expand Down Expand Up @@ -112,12 +116,12 @@ cdef void w2v_fast_sentence_sg_hs(
"""

cdef long long a, b
cdef long long row1 = word2_index * size, row2, sgn
cdef long long row1 = _mul(word2_index, size), row2, sgn
cdef REAL_t f, g, f_dot, lprob

memset(work, 0, size * cython.sizeof(REAL_t))
for b in range(codelen):
row2 = word_point[b] * size
row2 = _mul(word_point[b], size)
f_dot = our_dot(&size, &syn0[row1], &ONE, &syn1[row2], &ONE)
if f_dot <= -MAX_EXP or f_dot >= MAX_EXP:
continue
Expand Down Expand Up @@ -206,7 +210,7 @@ cdef unsigned long long w2v_fast_sentence_sg_neg(

"""
cdef long long a
cdef long long row1 = word2_index * size, row2
cdef long long row1 = _mul(word2_index, size), row2
cdef unsigned long long modulo = 281474976710655ULL
cdef REAL_t f, g, label, f_dot, log_e_f_dot
cdef np.uint32_t target_index
Expand All @@ -225,7 +229,7 @@ cdef unsigned long long w2v_fast_sentence_sg_neg(
continue
label = <REAL_t>0.0

row2 = target_index * size
row2 = _mul(target_index, size)
f_dot = our_dot(&size, &syn0[row1], &ONE, &syn1neg[row2], &ONE)
if f_dot <= -MAX_EXP or f_dot >= MAX_EXP:
continue
Expand Down Expand Up @@ -309,15 +313,15 @@ cdef void w2v_fast_sentence_cbow_hs(
continue
else:
count += ONEF
our_saxpy(&size, &ONEF, &syn0[indexes[m] * size], &ONE, neu1, &ONE)
our_saxpy(&size, &ONEF, &syn0[_mul(indexes[m], size)], &ONE, neu1, &ONE)
if count > (<REAL_t>0.5):
inv_count = ONEF/count
if cbow_mean:
sscal(&size, &inv_count, neu1, &ONE) # (does this need BLAS-variants like saxpy?)

memset(work, 0, size * cython.sizeof(REAL_t))
for b in range(codelens[i]):
row2 = word_point[b] * size
row2 = _mul(word_point[b], size)
f_dot = our_dot(&size, neu1, &ONE, &syn1[row2], &ONE)
if f_dot <= -MAX_EXP or f_dot >= MAX_EXP:
continue
Expand All @@ -342,7 +346,7 @@ cdef void w2v_fast_sentence_cbow_hs(
if m == i:
continue
else:
our_saxpy(&size, &words_lockf[indexes[m] % lockf_len], work, &ONE, &syn0[indexes[m] * size], &ONE)
our_saxpy(&size, &words_lockf[indexes[m] % lockf_len], work, &ONE, &syn0[_mul(indexes[m],size)], &ONE)


cdef unsigned long long w2v_fast_sentence_cbow_neg(
Expand Down Expand Up @@ -416,7 +420,7 @@ cdef unsigned long long w2v_fast_sentence_cbow_neg(
continue
else:
count += ONEF
our_saxpy(&size, &ONEF, &syn0[indexes[m] * size], &ONE, neu1, &ONE)
our_saxpy(&size, &ONEF, &syn0[_mul(indexes[m], size)], &ONE, neu1, &ONE)
if count > (<REAL_t>0.5):
inv_count = ONEF/count
if cbow_mean:
Expand All @@ -435,7 +439,7 @@ cdef unsigned long long w2v_fast_sentence_cbow_neg(
continue
label = <REAL_t>0.0

row2 = target_index * size
row2 = _mul(target_index, size)
f_dot = our_dot(&size, neu1, &ONE, &syn1neg[row2], &ONE)
if f_dot <= -MAX_EXP or f_dot >= MAX_EXP:
continue
Expand All @@ -459,7 +463,7 @@ cdef unsigned long long w2v_fast_sentence_cbow_neg(
if m == i:
continue
else:
our_saxpy(&size, &words_lockf[indexes[m] % lockf_len], work, &ONE, &syn0[indexes[m]*size], &ONE)
our_saxpy(&size, &words_lockf[indexes[m] % lockf_len], work, &ONE, &syn0[_mul(indexes[m], size)], &ONE)

return next_random

Expand Down Expand Up @@ -784,11 +788,11 @@ cdef void score_pair_sg_hs(
const np.uint32_t word2_index, REAL_t *work) nogil:

cdef long long b
cdef long long row1 = word2_index * size, row2, sgn
cdef long long row1 = _mul(word2_index, size), row2, sgn
cdef REAL_t f

for b in range(codelen):
row2 = word_point[b] * size
row2 = _mul(word_point[b], size)
f = our_dot(&size, &syn0[row1], &ONE, &syn1[row2], &ONE)
sgn = (-1)**word_code[b] # ch function: 0-> 1, 1 -> -1
f *= sgn
Expand Down Expand Up @@ -889,14 +893,14 @@ cdef void score_pair_cbow_hs(
continue
else:
count += ONEF
our_saxpy(&size, &ONEF, &syn0[indexes[m] * size], &ONE, neu1, &ONE)
our_saxpy(&size, &ONEF, &syn0[_mul(indexes[m], size)], &ONE, neu1, &ONE)
if count > (<REAL_t>0.5):
inv_count = ONEF/count
if cbow_mean:
sscal(&size, &inv_count, neu1, &ONE)

for b in range(codelens[i]):
row2 = word_point[b] * size
row2 = _mul(word_point[b], size)
f = our_dot(&size, neu1, &ONE, &syn1[row2], &ONE)
sgn = (-1)**word_code[b] # ch function: 0-> 1, 1 -> -1
f *= sgn
Expand Down