fq-hll 1.0.1

A Frequency-Quantized HyperLogLog Autocorrection algorithm that also is dyslexia-friendly.

FQ-HLL

An improved Frequency-Quantized HyperLogLog (FQ-HLL) Autocorrection Library

For theoretical value, there is nothing in my algorithm that uses any information about the English language or the QWERTY keyboard, to do any of the corrections. It is a NON-ML algorithm too.

Usage

The library defaults to searching within its own folder before searching in your local directory. Visit my GitHub Repo for more details, but there are two text files offered as base dictionaries: 20k_shun4midx.txt and database.txt, with around 20000 and 400 words respectively. The below code would only visit the local directory. If no dictionary is specified, 20k_shun4midx.txt would be used instead.

# ======== SAMPLE USAGE ======== #
if __name__ == "__main__":
    ac = Autocorrector()

    # File
    ans1 = ac.autocorrect("test_files/typo_file.txt", "test_files/class_suggestions.txt")
    print(ans1)

    ans2 = ac.top3("test_files/typo_file.txt", "test_files/class_suggestions.txt")

    # Optionally, you can not want it to output it into a file, then:
    # Individual strings
    ans3 = ac.autocorrect("hillo")
    ans4 = ac.top3("hillo")

    # Arrays
    ans5 = ac.autocorrect(["tsetign", "hillo", "goobye", "haedhpoesn"])
    ans6 = ac.top3(["tsetign", "hillo", "goobye", "haedhpoesn"])

    # You can even have a custom dictionary!
    dictionary = ["apple", "banana", "grape", "orange"]
    custom_ac = Autocorrector(dictionary)

    ans7 = custom_ac.autocorrect(["applle", "banana", "banan", "orenge", "grap", "pineapple"])
    ans8 = custom_ac.top3(["applle", "banana", "banan", "orenge", "grap", "pineapple"])

    print(ans7)
    print(ans8)

Results

More detail can be found in my GitHub Repo.

TL;DR it is way more accurate than traditional non-ML, non-language-specific algorithms such as BK-Tree and SymSpell and it also quite fast.

Notes

• HLL naturally doesn't have Local Differential Privacy (LDP) yet, but has natural obfuscation.
• This library does not collect personal data. However, still use it at your own discretion.

Dyslexia

Personally, I've always had an interest in autocorrect because I'm dyslexic and often unintentionally scramble or reverse letters when I read. Here are my thoughts about this algorithm based on my dyslexia.

• Reasoning would be more detailed in the algo_description/description.pdf file, but I find this algorithm's autocorrection suggestions are sometimes more intuitive (e.g. "klof" -> "folk") to my dyslexia than other Levenshtein distance-based autocorrection models.
• As a side note, as a dyslexic person, I naturally process words similar to how the FQ-HLL algorithm processes words, and that was my intuition in terms of how to create this algorithm in the first place.