Comments (3)
@jayseok-park Are there any benchmarks to define these constrains?
from dps.
There is a report on validation loss at a smaller scale (1.4B) of quality filtering, but there is no ablation for each constraints. I have no idea how they defined these numbers.
https://arxiv.org/abs/2112.11446
or maybe for now, it might be ok to add a filter like this and adjust the numbers later by looking at the distribution or something like data exploration.
from dps.
Re-consider about filter functions
doc_len_filter
: Filter any doc that does not contain between min_doc_len and max_doc_len words- This function is necessary to filter text dataset
mean_word_len_filter
: Filter any doc whose mean word length is outside the range of min_word_len to max_word_len characters- In Korean language, we do understand text without space, so this may not be needed
symbol_to_word_ratio_filter
: Filter any doc with a symbol-to-word ratio greater than symbol_to_word_ratio for either the hash symbol or the ellipsis- Can follow current defined setting
bullet_ellipsis_filter
: Filter any doc with more than bullet_point_ratio of lines starting with a bullet point, or more than ellipsis_ratio ending with an ellipsis- Can follow current defined setting
alphabetic_word_ratio_filter
: Filter any doc that alphabetic_word_ratio of words in a document does not contain at least one alphabetic character- Can follow current defined setting
least_k_essential_words_filter
: Filter any doc that does not contain at least k of the following English word_list: the, be, to, of, and, that, have, with (language specific words may needed)- This function needs to discussion to pick k-words
from dps.
Related Issues (20)
- Add normalize `?,:"!` in common preprocess job
- Update additional preprocess function HOT 1
- Remove `soynlp` library
- Add pre-processing for Japanese texts
- Replace html2text from Beautifulsoup HOT 1
- Task consideration HOT 3
- Implement minhash dedup module
- [ja] replace Japanese PII
- [ja] reduce emoticon HOT 1
- [ja] spam word filter
- Japanese pre-procesesing - remove text with low rate of Japanese stopwords HOT 4
- Improve Korean preprocessing algorithm
- Need to add ignore null or empty text during korean text process
- Refactor RDD process to Dataframe process
- [ja] refactor MinHashLSH-based near deduplication method
- Chiese dedup memory error HOT 1
- [ja] `.filter` is used instead of `.map` for non-filter methods HOT 1
- Bug in the function `remove_repeated_text`
- dedup_job java.lang.UnsatisfiedLinkError HOT 3
- k
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from dps.