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

Fixed bug where in masked_lm transformations only subwords were candidates for top_words #417

Merged
merged 19 commits into from
Feb 15, 2021
Merged

Fixed bug where in masked_lm transformations only subwords were candidates for top_words #417

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
19 commits
Select commit Hold shift + click to select a range
b0a9c97
Fixed bug where only subwords were candidates for top_words
Feb 10, 2021
3828c74
Update README.md
tk-dev11 Feb 11, 2021
4316c02
update test output to ignore num_queries
jinyongyoo Feb 12, 2021
08302ea
update test
jinyongyoo Feb 12, 2021
50daaa9
update test to fix inconsistency introduced by new stanza resource
jinyongyoo Feb 12, 2021
261c026
Merge pull request #421 from QData/fix-test
jinyongyoo Feb 12, 2021
aa2ccb5
Update README.md
qiyanjun Feb 13, 2021
4019eb1
reduce font of Model/readme.md
qiyanjun Feb 13, 2021
29bf26f
make all url smaller font in models/readme.md
qiyanjun Feb 13, 2021
e32476a
fix minor errors
qiyanjun Feb 14, 2021
20f79e1
add more details in the model page on readthedoc
qiyanjun Feb 14, 2021
47a03e3
add in sphinx-markdown-tables
qiyanjun Feb 14, 2021
2009a77
Merge pull request #419 from tahmid-kazi/patch-4
qiyanjun Feb 14, 2021
2884cbf
sphinx-markdown-tables runs fine on readthedoc , but not in Github Wo…
qiyanjun Feb 14, 2021
91e364a
bypass the sphinx issue on ubuntu-latest to ubuntu-18.04
qiyanjun Feb 14, 2021
c2bc09f
Revert "bypass the sphinx issue on ubuntu-latest to ubuntu-18.04"
qiyanjun Feb 14, 2021
5780b17
Update requirements.txt
qiyanjun Feb 14, 2021
86a3d31
Fixed bug where only subwords were candidates for top_words
Feb 10, 2021
203dba9
Merge remote-tracking branch 'origin/fix_transformation_masked_lm' in…
Feb 15, 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
80 changes: 79 additions & 1 deletion docs/3recipes/models.md
View file
Open in desktop

Large diffs are not rendered by default.

1 change: 1 addition & 0 deletions docs/conf.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,6 +43,7 @@
"nbsphinx",
# Enable .md doc files
"recommonmark",
"sphinx_markdown_tables",
]
autosummary_generate = True

Expand Down
1 change: 1 addition & 0 deletions docs/requirements.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,5 @@
recommonmark
sphinx-markdown-tables
nbsphinx
sphinx - autobuild
sphinx - rtd - theme
1 change: 1 addition & 0 deletions requirements.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,3 +18,4 @@ word2number
num2words
more-itertools
PySocks!=1.5.7,>=1.5.6
sphinx-markdown-tables
6 changes: 3 additions & 3 deletions tests/sample_outputs/run_attack_stanza_pos_tagger.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,11 +26,11 @@ lovingly photographed in the manner of a golden book sprung to subsistence


--------------------------------------------- Result 2 ---------------------------------------------
Positive (99%) --> Negative (58%)
Positive (99%) --> Negative (91%)

consistently clever and suspenseful .

persistently brainy and suspenseful .
persistently malin and suspenseful .


--------------------------------------------- Result 3 ---------------------------------------------
Expand Down Expand Up @@ -59,5 +59,5 @@ the story gives ample opportunity for large-scale action and suspense , which di
| Attack success rate: | 100.0% |
| Average perturbed word %: | 22.04% |
| Average num. words per input: | 15.5 |
| Avg num queries: | 175.67 |
| Avg num queries: | 174.67 |
+-------------------------------+--------+
77 changes: 77 additions & 0 deletions textattack/models/README.md
View file
Open in desktop

Large diffs are not rendered by default.

2 changes: 1 addition & 1 deletion textattack/shared/utils/install.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -139,7 +139,7 @@ def set_cache_dir(cache_dir):
def _post_install_if_needed():
"""Runs _post_install if hasn't been run since install."""
# Check for post-install file.
post_install_file_path = path_in_cache("post_install_check_2")
post_install_file_path = path_in_cache("post_install_check_3")
post_install_file_lock_path = post_install_file_path + ".lock"
post_install_file_lock = filelock.FileLock(post_install_file_lock_path)
post_install_file_lock.acquire()
Expand Down
18 changes: 9 additions & 9 deletions textattack/transformations/word_insertions/word_insertion_masked_lm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -113,21 +113,21 @@ def _get_new_words(self, current_text, indices_to_modify):
top_words = []
for _id in ranked_indices:
_id = _id.item()
token = self._lm_tokenizer.convert_ids_to_tokens(_id)
word = self._lm_tokenizer.convert_ids_to_tokens(_id)
if utils.check_if_subword(
token,
word,
self._language_model.config.model_type,
(masked_index == 1),
):
word = utils.strip_BPE_artifacts(
token, self._language_model.config.model_type
word, self._language_model.config.model_type
)
if (
mask_token_probs[_id] >= self.min_confidence
and utils.is_one_word(word)
and not utils.check_if_punctuations(word)
):
top_words.append(word)
if (
mask_token_probs[_id] >= self.min_confidence
and utils.is_one_word(word)
and not utils.check_if_punctuations(word)
):
top_words.append(word)

if (
len(top_words) >= self.max_candidates
Expand Down
18 changes: 9 additions & 9 deletions textattack/transformations/word_merges/word_merge_masked_lm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -112,21 +112,21 @@ def _get_merged_words(self, current_text, indices_to_modify):
top_words = []
for _id in ranked_indices:
_id = _id.item()
token = self._lm_tokenizer.convert_ids_to_tokens(_id)
word = self._lm_tokenizer.convert_ids_to_tokens(_id)
if utils.check_if_subword(
token,
word,
self._language_model.config.model_type,
(masked_index == 1),
):
word = utils.strip_BPE_artifacts(
token, self._language_model.config.model_type
word, self._language_model.config.model_type
)
if (
mask_token_probs[_id] >= self.min_confidence
and utils.is_one_word(word)
and not utils.check_if_punctuations(word)
):
top_words.append(word)
if (
mask_token_probs[_id] >= self.min_confidence
and utils.is_one_word(word)
and not utils.check_if_punctuations(word)
):
top_words.append(word)

if (
len(top_words) >= self.max_candidates
Expand Down
18 changes: 9 additions & 9 deletions textattack/transformations/word_swaps/word_swap_masked_lm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -136,21 +136,21 @@ def _bae_replacement_words(self, current_text, indices_to_modify):
top_words = []
for _id in ranked_indices:
_id = _id.item()
token = self._lm_tokenizer.convert_ids_to_tokens(_id)
word = self._lm_tokenizer.convert_ids_to_tokens(_id)
if utils.check_if_subword(
token,
word,
self._language_model.config.model_type,
(masked_index == 1),
):
word = utils.strip_BPE_artifacts(
token, self._language_model.config.model_type
word, self._language_model.config.model_type
)
if (
mask_token_probs[_id] >= self.min_confidence
and utils.is_one_word(word)
and not utils.check_if_punctuations(word)
):
top_words.append(word)
if (
mask_token_probs[_id] >= self.min_confidence
and utils.is_one_word(word)
and not utils.check_if_punctuations(word)
):
top_words.append(word)

if (
len(top_words) >= self.max_candidates
Expand Down