Skip to content

Latest commit

 

History

History
83 lines (64 loc) · 4.45 KB

README.md

File metadata and controls

83 lines (64 loc) · 4.45 KB
Raw

Document Quality Transform

Please see the set of transform project conventions for details on general project conventions, transform configuration, testing and IDE set up.

Summary

This transform will calculate and annotate several metrics related to document, which are usuful to see the quality of document.

In this transform, following metrics will be included:

output column name data type description supported language
docq_total_words int the total number of words ALL
docq_mean_word_len int the mean of words' lengths ALL
docq_symbol_to_word_ratio float the ratio of symbol-to-word ratio (Reference for symbols like emojis: https://textacy.readthedocs.io/en/0.11.0/api_reference/preprocessing.html, currently used symbol: #, ...) ALL
docq_sentence_count int the number of sentences ALL
docq_curly_bracket_ratio float the ratio between the number of occurrences of { or } over the text length ALL
docq_lorem_ipsum_ratio float the ratio between the number of occurrences of lorem ipsum over the text length. Lorem ipsum, or lipsum as it is sometimes known, is dummy text used in laying out print, graphic or web designs. ALL
docq_contain_bad_word bool whether text containst bad words ALL
docq_bullet_point_ratio float the ratio of lines starting with a bullet point ALL
docq_ellipsis_line_ratio float the ratio of lines ending with an ellipsis ALL
docq_alphabet_word_ratio float the ratio of words having at least one alphabetic character ALL
docq_contain_common_en_words bool whether the given text contains common English words like the, and, to, that, of, with, be, and have ALL
docq_avg_ja_sentence_len int average sentence length for an input text, inspired by an OSS HojiChar. ja
docq_first_ja_alphabet_pos int first position of occurrence of Japanese alphabets (i.e., Hiragana or Katakana) ja

You can see more detailed backgrounds of some columns in Deepmind's Gopher paper

Configuration and command line Options

The set of dictionary keys holding DocQualityTransform configuration for values are as follows:

  • text_lang - specifies language used in the text content. By default, "en" is used.
  • doc_content_column - specifies column name that contains document text. By default, "contents" is used.
  • bad_word_filepath - specifies a path to bad word file: local folder (file or directory) that points to bad word file. You don't have to set this parameter if you don't need to set bad words.

Running

Launched Command Line Options

When running the transform with the Ray launcher (i.e. TransformLauncher), the following command line arguments are available in addition to the options provided by the python launcher.

  --docq_text_lang DOCQ_TEXT_LANG   language used in the text content. By default, "en" is used.
  --docq_doc_content_column DOCQ_DOC_CONTENT_COLUMN   column name that contain document text. By default, "contents" is used.
  --docq_bad_word_filepath DOCQ_BAD_WORD_FILEPATH   path to bad word file: local folder (file or directory) that points to bad word file. You don't have to set this parameter if you don't need to set bad words.

These correspond to the configuration keys described above.

Running the samples

To run the samples, use the following make targets

  • run-cli-sample - runs src/doc_quality_transform.py using command line args
  • run-local-sample - runs src/doc_quality_local.py

These targets will activate the virtual environment and set up any configuration needed. Use the -n option of make to see the detail of what is done to run the sample.

For example,

make run-cli-sample
...

Then

ls output

To see results of the transform.

Transforming data using the transform image

To use the transform image to transform your data, please refer to the running images quickstart, substituting the name of this transform image and runtime as appropriate.

Troubleshooting guide

For M1 Mac user, if you see following error during make command, error: command '/usr/bin/clang' failed with exit code 1, you may better follow this step