matcher-py 0.15.5

A high-performance matcher designed to solve LOGICAL and TEXT VARIATIONS problems in word matching, implemented in Rust.

Matcher Python Bindings (PyO3)

Python bindings for the Matcher library — a high-performance matcher designed to solve LOGICAL and TEXT VARIATIONS problems in word matching, implemented in Rust via PyO3.

For detailed implementation, see the Design Document.

Features

• Text Transformation:
  • VariantNorm: CJK variant normalization. Example: 測試 -> 测试
  • Delete: Remove specific characters. Example: *Fu&*iii&^%%*&kkkk -> Fuiiikkkk
  • Normalize: Normalize special characters to identifiable characters. Example: ＡＢⅣ①℉ -> ab41°f
  • Romanize: Convert CJK characters to space-separated romanized form (Pinyin, Romaji, RR) for fuzzy matching. Example: 西安 -> xi an, matches 洗按 -> xi an, but not 先 -> xian
  • RomanizeChar: Convert CJK characters to romanized form without boundary spaces. Example: 西安 -> xian, matches 洗按 and 先 -> xian
  • EmojiNorm: Convert emoji to English words (CLDR short names) and strip modifiers. Example: 👍🏽 -> thumbs_up, 🔥 -> fire
• AND OR NOT Word Matching:
  • Takes into account the number of repetitions of words.
  • & (AND): hello&world matches hello world and world,hello
  • | (OR): color|colour matches color and colour
  • ~ (NOT): hello~helloo~hhello matches hello but not helloo and hhello
  • \b (word boundary): \bcat\b matches "the cat" but not "concatenate"
  • Repeated segments: 无&法&无&天 matches 无无法天 (because 无 is repeated twice), but not 无法天
  • Combined: color|colour&bright~dark matches "bright color" but not "dark colour"
• Pickle Support: SimpleMatcher instances can be pickled and unpickled for serialization.

Installation

Use pip

pip install matcher_py

Build from source

Requires the Rust nightly toolchain.

git clone https://github.com/Lips7/Matcher.git
cd Matcher/matcher_py

# Option 1: Using uv (recommended for development)
pip install uv
uv sync

# Option 2: Using maturin directly
pip install maturin
maturin develop --release

Usage

All relevant types are defined in matcher_py.pyi.

Text Process Usage

Here’s an example of how to use the reduce_text_process and text_process functions:

from matcher_py import ProcessType, reduce_text_process, text_process

# Combine and reduce multiple transformations
print(reduce_text_process(ProcessType.DELETE_NORMALIZE, "hello, world!"))
# Perform a single transformation
print(text_process(ProcessType.DELETE, "hello, world!"))

Simple Matcher — Builder (recommended)

from matcher_py import ProcessType, SimpleMatcherBuilder

builder = SimpleMatcherBuilder()
builder.add_word(ProcessType.NONE, 1, "hello&world")
builder.add_word(ProcessType.NONE, 2, "word&word~hello")
builder.add_word(ProcessType.DELETE, 3, "hallo")
matcher = builder.build()

assert matcher.is_match("hello^&!#*#&!^#*()world")
result = matcher.process("hello,world,word,word,hallo")
print(result)

Simple Matcher — JSON constructor

You can also construct from JSON bytes if you already have them:

import json

from matcher_py import ProcessType, SimpleMatcher

matcher = SimpleMatcher(
    json.dumps(
        {
            ProcessType.NONE: {
                1: "hello&world",
                2: "word&word~hello"
            },
            ProcessType.DELETE: {
                3: "hallo"
            }
        }
    ).encode()
)
assert matcher.is_match("hello^&!#*#&!^#*()world")

OR, NOT, and Word Boundary

from matcher_py import ProcessType, SimpleMatcherBuilder

builder = SimpleMatcherBuilder()
builder.add_word(ProcessType.NONE, 1, "color|colour")                     # OR
builder.add_word(ProcessType.NONE, 2, "banana~peel")                      # NOT
builder.add_word(ProcessType.NONE, 3, r"\bcat\b")                         # word boundary
builder.add_word(ProcessType.NONE, 4, r"bright&color|colour~\bdark\b")    # combined
matcher = builder.build()

assert matcher.is_match("nice colour")            # OR
assert matcher.is_match("banana split")            # NOT (no veto)
assert not matcher.is_match("banana peel")         # NOT (vetoed)
assert matcher.is_match("the cat sat")             # word boundary
assert not matcher.is_match("concatenate")         # "cat" is substring
assert matcher.is_match("bright colour")           # combined: AND + OR
assert not matcher.is_match("bright dark color")   # combined: vetoed by \bdark\b
assert matcher.is_match("bright darken color")     # "darken" ≠ \bdark\b

Explanation of the configuration

• SimpleMatcher's configuration is defined by the SimpleTable = Dict[ProcessType, Dict[int, str]] type, the value Dict[int, str]'s key is called word_id, word_id is required to be globally unique.

ProcessType

• NONE: No transformation.
• VARIANT_NORM: CJK variant normalization. Based on VARIANT_NORM.
  • 測試 -> 测试
  • 現⾝ -> 现身
• DELETE: Delete all punctuation, special characters, separator characters, and configured control/format codepoints. Based on TEXT_DELETE.
  • hello, world! -> helloworld
  • 《你∷好》 -> 你好
• NORMALIZE: Normalize all English character variations and number variations to basic characters. Based on NORM and NUM_NORM.
  • ＡＢⅣ①℉ -> ab41°f
  • ⅠⅡⅢ -> 123
• ROMANIZE: Convert CJK characters to space-separated romanization (Pinyin, Romaji, RR). Based on ROMANIZE.
  • 你好 -> ni hao
  • 西安 -> xi an
• ROMANIZE_CHAR: Convert CJK characters to romanized form without boundary spaces. Based on ROMANIZE.
  • 你好 -> nihao
  • 西安 -> xian
• EMOJI_NORM: Convert emoji to English words (CLDR short names) and strip modifiers. Based on EMOJI_NORM.
  • 👍🏽 -> thumbs_up
  • 🔥 -> fire

You can combine these transformations as needed. Pre-defined combinations like DELETE_NORMALIZE and VARIANT_NORM_DELETE_NORMALIZE are provided for convenience.

Be careful combining ROMANIZE and ROMANIZE_CHAR: they preserve different word boundaries, so the same input can behave like xi + an in one pipeline and xian in the other.

Error Handling

• Construction (SimpleMatcher(bytes)): raises ValueError if the JSON is malformed or contains invalid ProcessType values. This is the only operation that can fail.
• Matching (is_match, process, batch_*): infallible once the matcher is built. These methods never raise exceptions.

Contributing

Contributions to matcher_py are welcome! If you find a bug or have a feature request, please open an issue on the GitHub repository. If you would like to contribute code, please fork the repository and submit a pull request.

License

matcher_py is licensed under the MIT OR Apache-2.0 license.

More Information

For more details, visit the GitHub repository.