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.

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

add abbreviation replacement data augmentation op and test #732

Open
wants to merge 16 commits into
base: master
Choose a base branch
from

Conversation

abbeyyyy
Copy link
Collaborator

@abbeyyyy abbeyyyy commented Apr 9, 2022

This PR fixes #736.

Description of changes

Add Abbreviation Replacement Augmentation Method

Possible influences of this PR.

This PR provides a new replacement-based data augmentation method

Test Conducted

Test cases included in abbreviation_replacement_op_test.py

@hunterhector
Copy link
Member

hunterhector commented Apr 10, 2022

A couple of preparations for the PR:

  1. We will need to create an issue and associate a PR with the issue: https://github.com/asyml/forte/blob/master/CONTRIBUTING.md#pull-requests
  2. We've send you an invitation for this repo, then you can run the workflow CI without approval

@codecov
Copy link

codecov bot commented Apr 11, 2022

Codecov Report

Merging #732 (82da6ef) into master (61c44ac) will increase coverage by 0.03%.
The diff coverage is 91.80%.

@@            Coverage Diff             @@
##           master     #732      +/-   ##
==========================================
+ Coverage   80.94%   80.98%   +0.03%     
==========================================
  Files         249      251       +2     
  Lines       18664    18725      +61     
==========================================
+ Hits        15108    15164      +56     
- Misses       3556     3561       +5     
Impacted Files Coverage Δ
..._augment/algorithms/abbreviation_replacement_op.py 85.71% <85.71%> (ø)
...ent/algorithms/abbreviation_replacement_op_test.py 96.96% <96.96%> (ø)

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 61c44ac...82da6ef. Read the comment docs.

@abbeyyyy abbeyyyy marked this pull request as ready for review April 12, 2022 00:17
@abbeyyyy abbeyyyy requested a review from Pushkar-Bhuse April 12, 2022 00:20

class AbbreviationReplacementOp(SingleAnnotationAugmentOp):
r"""
This class is a replacement op utilizing a pre-defined
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The docstring should be more comprehensive. This is what the user is going to see if they want to use this DA op.

@@ -0,0 +1,104 @@
# Copyright 2020 The Forte Authors. All Rights Reserved.
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

2022*

super().__init__(configs)
if "dict_path" in configs.keys():
self.dict_path = configs["dict_path"]
else:
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

An if-else loop is not needed here as you are already setting a default value in the default_configs

self, input_anno: Annotation
) -> Tuple[bool, str]:
r"""
This function replaces a word from an abbreviation dictionary.
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Again, we should add a better description of what this function will do.

# If the replacement does not happen, return False.
if random.random() > self.configs.prob:
return False, input_anno.text
if input_anno.text in self.data.keys():
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Since you are returning from the function if the program enters the earlier if statement, you dont need to add this if

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Also, I am not sure is this check (input_anno.text in self.data.keys()) is necessary.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I was thinking if the input phrase does not have a corresponding abbreviation, an error will occur.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

  1. When checking dict existence, use text in self.data, don't need to call the keys().
  2. Now we can see that the prob only applies to the annotation that has an abbreviation, which should probably be specified in the class docstring.

- dict_path: the `url` or the path to the pre-defined
abbreviation json file. The key is a word / phrase we want
to replace. The value is an abbreviated word of the
corresponding key.
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'd recommend adding the default value of dict_path in the docstring as well since this is what will be rendered in the documentation and it would be easier for users to see.

Copy link
Collaborator

@Pushkar-Bhuse Pushkar-Bhuse left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Your implementation of the Op seems fine but it looks like you might have not gotten the underlying intricacies of how to modify the SingleAnnotationAugmentOp for different type of annotations. Just take a closer look at that once.

A dictionary with the default config for this processor.
Following are the keys for this dictionary:
- prob: The probability of replacement,
should fall in [0, 1]. Default value is 0.1
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The default value below is 0.5. Make sure you check the documentation thoroughly.

data_pack = DataPack()
text = "see you later"
data_pack.set_text(text)
token = Token(data_pack, 0, len(text))
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The Token class is generally used for a single word. When annotating the whole sequence, you should use the Sentence class. Also, if your SingleAnnotationOp augments an annotation other than a Token, you must specify that in the default_configs. For your reference, look at the implementation of the BackTranslationOp and its test cases. Even that Op augments sentences.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We also have Document https://github.com/asyml/forte/blob/master/ft/onto/base_ontology.py#L136 for the whole article.

I know it is just a test case so it doesn't matter too much, but still worth noting.

augmented_data_pack = self.abre.perform_augmentation(data_pack)

augmented_token = list(
augmented_data_pack.get("ft.onto.base_ontology.Token")
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think you should take the comment above into consideration and rework your test cases accordingly.

Copy link
Collaborator

@Pushkar-Bhuse Pushkar-Bhuse left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think the changes look good.

abbreviation dictionary to replace word or phrase
with an abbreviation. The abbreviation dictionary can
be user-defined, we also provide a default dictionary.
`prob` indicates the probability of replacement.
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What does "probability of replacement" mean? For example, if prob is 0.4, is the replacement happen 40% of the case or the other way around. Does it mean that 40% of the words will be replaced, etc. Let's specify this clearly.

# If the replacement does not happen, return False.
if random.random() > self.configs.prob:
return False, input_anno.text
if input_anno.text in self.data.keys():
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

  1. When checking dict existence, use text in self.data, don't need to call the keys().
  2. Now we can see that the prob only applies to the annotation that has an abbreviation, which should probably be specified in the class docstring.

if random.random() > self.configs.prob:
return False, input_anno.text
if input_anno.text in self.data.keys():
result: str = self.data[input_anno.text]
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Something about this replacement:

  1. Do we need to consider the case? Maybe we should lower case your dictionary and user input.
  2. How about substrings? For example, in "see you later": "syl8r", what if we have an input "i will see you later", it looks like we won't replace this?

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Maybe you need to consider using an Aho-Corasick data sturcture here: https://pyahocorasick.readthedocs.io/en/latest/

to replace. The value is an abbreviated word of the
corresponding key. Default dictionary is from a web-scraped
slang dictionary
("https://github.com/abbeyyyy/JsonFiles/blob/main/abbreviate.json").
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Did we adopt the dictionary from another library?

data_pack_1 = DataPack()
text_1 = "I will see you later!"
data_pack_1.set_text(text_1)
phrase_1 = Phrase(data_pack_1, 7, len(text_1) - 1)
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I see that you have to first identify the phrase before doing the match, which is not a very typical use case.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

Add abbreviation replacement data augmentation
4 participants