A Domain Specific Language (DSL) for building language patterns. These can be later compiled into spaCy patterns, pure regex, or any other format

Overview

Rita Logo

RITA DSL

Documentation Status codecov made-with-python Maintenance PyPI version fury.io PyPI download month GitHub license

This is a language, loosely based on language Apache UIMA RUTA, focused on writing manual language rules, which compiles into either spaCy compatible patterns, or pure regex. These patterns can be used for doing manual NER as well as used in other processes, like retokenizing and pure matching

An Introduction Video

Intro

Links

Support

reddit Gitter

Install

pip install rita-dsl

Simple Rules example

rules = """
cuts = {"fitted", "wide-cut"}
lengths = {"short", "long", "calf-length", "knee-length"}
fabric_types = {"soft", "airy", "crinkled"}
fabrics = {"velour", "chiffon", "knit", "woven", "stretch"}

{IN_LIST(cuts)?, IN_LIST(lengths), WORD("dress")}->MARK("DRESS_TYPE")
{IN_LIST(lengths), IN_LIST(cuts), WORD("dress")}->MARK("DRESS_TYPE")
{IN_LIST(fabric_types)?, IN_LIST(fabrics)}->MARK("DRESS_FABRIC")
"""

Loading in spaCy

import spacy
from rita.shortcuts import setup_spacy


nlp = spacy.load("en")
setup_spacy(nlp, rules_string=rules)

And using it:

>>> r = nlp("She was wearing a short wide-cut dress")
>>> [{"label": e.label_, "text": e.text} for e in r.ents]
[{'label': 'DRESS_TYPE', 'text': 'short wide-cut dress'}]

Loading using Regex (standalone)

import rita

patterns = rita.compile_string(rules, use_engine="standalone")

And using it:

>>> list(patterns.execute("She was wearing a short wide-cut dress"))
[{'end': 38, 'label': 'DRESS_TYPE', 'start': 18, 'text': 'short wide-cut dress'}]
Comments
  • Jetbrains RITA Plugin not compatible with PyCharm 2020.2.1

    Jetbrains RITA Plugin not compatible with PyCharm 2020.2.1

    Plugin Version: 1.2 https://plugins.jetbrains.com/plugin/15011-rita-language/versions/

    Tested Version: PyCharm 2020.2.1 (Professional Edition)

    Error when trying to install from disk: grafik

    On the plugin site https://plugins.jetbrains.com/plugin/15011-rita-language/versions/ it says, that this should be uncompitable for all IntellJ-based IDEs in the 2020.2 version:

    The list of supported products was determined by dependencies defined in the plugin.xml: Android Studio — build 201.7223 — 201.* DataGrip — 2020.1.3 — 2020.1.5 IntelliJ IDEA Ultimate — 2020.1.1 — 2020.1.4 Rider — 2020.1.3 PyCharm Professional — 2020.1.1 — 2020.1.4 PyCharm Community — 2020.1.1 — 2020.1.4 PhpStorm — 2020.1.1 — 2020.1.4 IntelliJ IDEA Educational — 2020.1.1 — 2020.1.2 CLion — 2020.1.1 — 2020.1.3 PyCharm Educational — 2020.1.1 — 2020.1.2 GoLand — 2020.1.1 — 2020.1.4 AppCode — 2020.1.2 — 2020.1.6 RubyMine — 2020.1.1 — 2020.1.4 MPS — 2020.1.1 — 2020.1.4 IntelliJ IDEA Community — 2020.1.1 — 2020.1.4 WebStorm — 2020.1.1 — 2020.1.4

    opened by rolandmueller 3
  • IN_LIST ignores OP quantifier

    IN_LIST ignores OP quantifier

    Somehow I get this unexpected behaviour when using OP quantifiers (?, *, +, etc) with the IN_LIST element:

    rules = """
    list_elements = {"one", "two"}
    {IN_LIST(list_elements)?}->MARK("LABEL")
    """
    rules = rita.compile_string(rules)
    expected_result = "[{'label': 'LABEL', 'pattern': [{'LOWER': {'REGEX': '^(one|two)$'}, 'OP': '?'}]}]"
    print("expected_result:", expected_result)
    print("result:", rules)
    assert str(rules) == expected_result
    

    Version: 0.5.0

    bug 
    opened by rolandmueller 3
  • Add module regex

    Add module regex

    This feature would introduce the REGEX element as a module.

    Matches words based on a Regex pattern e.g. all words that start with an 'a' would be REGEX("^a")

    !IMPORT("rita.modules.regex")
    
    {REGEX("^a")}->MARK("TAGGED_MATCH")
    
    opened by rolandmueller 2
  • Feature/pluralize

    Feature/pluralize

    Add a new module for a PLURALIZE tag For a noun or a list of nouns, it will match any singular or plural word. Usage for a single word, e.g.:

    PLURALIZE("car")
    

    Usage for lists, e.g.:

    vehicles = {"car", "bicycle", "ship"}
    PLURALIZE(vehicles)
    

    Will work even for regex or if the lemmatizer of spaCy is making an error. Has dependency to the Python inflect package https://pypi.org/project/inflect/

    opened by rolandmueller 2
  • Feature/regex tag

    Feature/regex tag

    This feature would introduce the TAG element as a module. Needs a new parser for the SpaCy translate. Would allow more flexible matching of detailed part-of-speech tag, like all adjectives or nouns: TAG("^NN|^JJ").

    opened by rolandmueller 2
  • Feature/improve robustness

    Feature/improve robustness

    In general - measure how long it takes to compile and avoid situations when pattern creates infinite loop (possible to get to this situation using regex).

    Closes: https://github.com/zaibacu/rita-dsl/issues/78

    opened by zaibacu 1
  • Add TAG_WORD macro to Tag module

    Add TAG_WORD macro to Tag module

    This feature would introduce the TAG_WORD element to the Tag module

    TAG_WORD is for generating TAG patterns with a word or a list.

    e.g. match only "proposed" when it is in the sentence a verb (and not an adjective):

    !IMPORT("rita.modules.tag")
    
    TAG_WORD("^VB", "proposed")
    

    or e.g. match a list of words only to verbs

    !IMPORT("rita.modules.tag")
    
    words = {"percived", "proposed"}
    {TAG_WORD("^VB", words)}->MARK("LABEL")
    
    opened by rolandmueller 1
  • Add Orth module

    Add Orth module

    This feature would introduce the ORTH element as a module.

    Ignores case-insensitive configuration and checks words as written that means case-sensitive even if configuration is case-insensitive. Especially useful for acronyms and proper names.

    Works only with spaCy engine

    Usage:

    !IMPORT("rita.modules.orth")
    
    {ORTH("IEEE")}->MARK("TAGGED_MATCH")
    
    opened by rolandmueller 1
  • Add conifugration for implicit hyphon characters between words

    Add conifugration for implicit hyphon characters between words

    Add a new Configuration implicit_hyphon (default false) for automatically adding hyphon characters - to the rules. Enabling implicit_hyphon is disabling implicit_punct. Rationale: implicit_punct is often to much inclusive. The implicit_punct has the hyphon token included, but it is adding (at least in my use case) unwanted tokens (like parentheses) to the matches, especially for more complex rules. So implicit_hyphon is a little bit more strict than implicit_punct.

    opened by rolandmueller 1
  • Fix sequencial optional

    Fix sequencial optional

    Closes https://github.com/zaibacu/rita-dsl/issues/69

    Turns out it is a bug related to - character which in most cases used as a splitter, but in this case as a stand alone word

    opened by zaibacu 1
  • Method to validate syntax

    Method to validate syntax

    Currently it can be partially done:

    from rita.parser import RitaParser
    from rita.config import SessionConfig
    config = SessionConfig()
    p = RitaParser(config)
    p.build()
    result = p.parse(rules)
    if result is None:
        raise RuntimeError("... Something is wrong with syntax")
    

    But it would be nice to have single method for that and have actual error info.

    enhancement 
    opened by zaibacu 0
  • Dynamic case sensitivity for Standalone Engine

    Dynamic case sensitivity for Standalone Engine

    We want to be able to make specified word inside pattern to be case sensitive, while rest of the pattern is case insensitive.

    It looks like it can be achieved using inline modifier groups regex feature, it requires Python3.6+ version

    enhancement 
    opened by zaibacu 0
  • JS rule engine

    JS rule engine

    Should work similarly to standalone engine, maybe even inherit most of it, but it should result into valid JavaScript code, preferably a single function to which you give raw text and get result of multiple parsed entities

    enhancement help wanted 
    opened by zaibacu 0
  • Allow LOAD macro to load from external locations

    Allow LOAD macro to load from external locations

    now LOAD(file_name) macro searches text file in current path.

    Usually reading from the local file is the best, but it should be cool, to be able just give like github GIST url and just load everything we need. This would be very useful for Demo page case

    good first issue 
    opened by zaibacu 0
Releases(0.7.0)
  • 0.7.0(Feb 2, 2021)

    0.7.0 (2021-02-02)


    Features

    • standalone engine now will return submatches list containing start and end for each part of match #93

    • Partially covered https://github.com/zaibacu/rita-dsl/issues/70

      Allow nested patterns, like:

          num_with_fractions = {NUM, WORD("-")?, IN_LIST(fractions)}
          complex_number = {NUM|PATTERN(num_with_fractions)}
    
          {PATTERN(complex_number)}->MARK("NUMBER")
    

    #95

    • Submatches for rita-rust engine #96

    • Regex module which allows to specify word pattern, eg. REGEX(^a) means word must start with letter "a"

      Implemented by: Roland M. Mueller (https://github.com/rolandmueller) #101

    • ORTH module which allows you to specify case sensitive entry while rest of the rules ignores case. Used for acronyms and proper names

      Implemented by: Roland M. Mueller (https://github.com/rolandmueller) #102

    • Additional macro for tag module, allowing to tag specific word/list of words

      Implemented by: Roland M. Mueller (https://github.com/rolandmueller) #103

    • Added names module which allows to generate person names variations #105

    • spaCy v3 Support #109

    Fix

    • Optimizations for Rust Engine

      • No need for passing text forward and backward, we can calculate from text[start:end]

      • Grouping and sorting logic can be done in binary code #88

    • Fix NUM parsing bug #90

    • Switch from (^\s) to \b when doing IN_LIST. Should solve several corner cases #91

    • Fix floating point number matching #92

    • revert #91 changes. Keep old way for word boundary #94

    Source code(tar.gz)
    Source code(zip)
  • 0.6.0(Aug 29, 2020)

    0.6.0 (2020-08-29)


    Features

    • Implemented ability to alias macros, eg.:
          numbers = {"one", "two", "three"}
          @alias IN_LIST IL
    
          IL(numbers) -> MARK("NUMBER")
    

    Now using "IL" will actually call "IN_LIST" macro. #66

    • introduce the TAG element as a module. Needs a new parser for the SpaCy translate. Would allow more flexible matching of detailed part-of-speech tag, like all adjectives or nouns: TAG("^NN|^JJ").

      Implemented by: Roland M. Mueller (https://github.com/rolandmueller) #81

    • Add a new module for a PLURALIZE tag For a noun or a list of nouns, it will match any singular or plural word.

      Implemented by: Roland M. Mueller (https://github.com/rolandmueller) #82

    • Add a new Configuration implicit_hyphon (default false) for automatically adding hyphon characters - to the rules.

      Implemented by: Roland M. Mueller (https://github.com/rolandmueller) #84

    • Allow to give custom regex impl. By default re is used #86

    • An interface to be able to use rust engine.

      In general it's identical to standalone, but differs in one crucial part - all of the rules are compiled into actual binary code and that provides large performance boost. It is proprietary, because there are various caveats, engine itself is a bit more fragile and needs to be tinkered to be optimized to very specific case (eg. few long texts with many matches vs a lot short texts with few matches). #87

    Fix

    • Fix - bug when it is used as stand alone word #71
    • Fix regex matching, when shortest word is selected from IN_LIST #72
    • Fix IN_LIST regex so that it wouldn't take part of word #75
    • Fix IN_LIST operation bug - it was ignoring them #77
    • Use list branching only when using spaCy Engine #80
    Source code(tar.gz)
    Source code(zip)
  • 0.5.0(Jun 18, 2020)

    Features

    • Added PREFIX macro which allows to attach word in front of list items or words #47

    • Allow to pass variables directly when doing compile and compile_string #51

    • Allow to compile (and later load) rules using rita CLI while using standalone engine (spacy is already supported) #53

    • Added ability to import rule files into rule file. Recursive import is supported as well. #55

    • Added possibility to define pattern as a variable and reuse it in other patterns:

      Example:

    ComplexNumber = {NUM+, WORD("/")?, NUM?}
    
    {PATTERN(ComplexNumber), WORD("inches"), WORD("Height")}->MARK("HEIGHT")
    {PATTERN(ComplexNumber), WORD("inches"), WORD("Width")}->MARK("WIDTH")
    

    #64

    Fix

    • Fix issue with multiple wildcard words using standalone engine #46
    • Don't crash when no rules are provided #50
    • Fix Number and ANY-OF parsing #59
    • Allow escape characters inside LITERAL #62
    Source code(tar.gz)
    Source code(zip)
  • 0.4.0(Jan 25, 2020)

    0.4.0 (2020-01-25)


    Features

    • Support for deaccent. In general, if accented version of word is given, both deaccented and accented will be used to match. To turn iit off - !CONFIG("deaccent", "N") #38
    • Added shortcuts module to simplify injecting into spaCy #42

    Fix

    • Fix issue regarding Spacy rules with IN_LIST and using case-sensitive mode. It was creating Regex pattern which is not valid spacy pattern #40
    Source code(tar.gz)
    Source code(zip)
  • 0.3.2(Dec 19, 2019)

    Features

      • Introduced towncrier to track changes
      • Added linter flake8
      • Refactored code to match pep8 #32

    Fix

      • Fix WORD split by -

      • Split by (empty space) as well

      • Coverage score increase #35

    Source code(tar.gz)
    Source code(zip)
  • 0.3.0(Dec 14, 2019)

    Now there's one global config and child config created per-session (one session = one rule file compilation). Imports and variables are stored in this config as well.

    Remove context argument from MACROS, making code cleaner and easier to read

    Source code(tar.gz)
    Source code(zip)
  • 0.2.2(Dec 8, 2019)

    Features of up to this point:

    • Standalone parser - can use internal regex rather than spaCy if you need to
    • Ability to do logical OR in rule. eg.: {WORD(w1)|WORD(w2),WORD(w3)} would result into two rules: {WORD(w1),WORD(w3)} and {WORD(w2),WORD(w3)}
    • Exclude operator {WORD(w1), WORD(w2)!} would match w1 and anything but w2
    Source code(tar.gz)
    Source code(zip)
Owner
Šarūnas Navickas
Data Engineer @ TokenMill. Doing BJJ @ Voras-Bjj. Dad @ Home.
Šarūnas Navickas
Use Tensorflow2.7.0 Build OpenAI'GPT-2

TF2_GPT-2 Use Tensorflow2.7.0 Build OpenAI'GPT-2 使用最新tensorflow2.7.0构建openai官方的GPT-2 NLP模型 优点 使用无监督技术 拥有大量词汇量 可实现续写(堪比“xx梦续写”) 实现对话后续将应用于FloatTech的Bot

Watermelon 9 Sep 13, 2022
Yes it's true :broken_heart:

Information WARNING: No longer hosted If you would like to be on this repo's readme simply fork or star it! Forks 1 - Flowzii 2 - Errorcrafter 3 - vk-

Dropout 66 Dec 31, 2022
Unsupervised Language Modeling at scale for robust sentiment classification

** DEPRECATED ** This repo has been deprecated. Please visit Megatron-LM for our up to date Large-scale unsupervised pretraining and finetuning code.

NVIDIA Corporation 1k Nov 17, 2022
Python api wrapper for JellyFish Lights

Python api wrapper for JellyFish Lights The hope is to make this a pip installable package Current capabalilities: Connects to a local JellyFish Light

10 Dec 18, 2022
Code repository of the paper Neural circuit policies enabling auditable autonomy published in Nature Machine Intelligence

Code repository of the paper Neural circuit policies enabling auditable autonomy published in Nature Machine Intelligence

9 Jan 08, 2023
Simplified diarization pipeline using some pretrained models - audio file to diarized segments in a few lines of code

simple_diarizer Simplified diarization pipeline using some pretrained models. Made to be a simple as possible to go from an input audio file to diariz

Chau 65 Dec 30, 2022
Python bindings to the dutch NLP tool Frog (pos tagger, lemmatiser, NER tagger, morphological analysis, shallow parser, dependency parser)

Frog for Python This is a Python binding to the Natural Language Processing suite Frog. Frog is intended for Dutch and performs part-of-speech tagging

Maarten van Gompel 46 Dec 14, 2022
End-to-end MLOps pipeline of a BERT model for emotion classification.

image source EmoBERT-MLOps The goal of this repository is to build an end-to-end MLOps pipeline based on the MLOps course from Made with ML, but this

Dimitre Oliveira 4 Nov 06, 2022
Conditional probing: measuring usable information beyond a baseline

Conditional probing: measuring usable information beyond a baseline

John Hewitt 20 Dec 15, 2022
Open-Source Toolkit for End-to-End Speech Recognition leveraging PyTorch-Lightning and Hydra.

🤗 Contributing to OpenSpeech 🤗 OpenSpeech provides reference implementations of various ASR modeling papers and three languages recipe to perform ta

Openspeech TEAM 513 Jan 03, 2023
OpenChat: Opensource chatting framework for generative models

OpenChat is opensource chatting framework for generative models.

Hyunwoong Ko 427 Jan 06, 2023
Natural language Understanding Toolkit

Natural language Understanding Toolkit TOC Requirements Installation Documentation CLSCL NER References Requirements To install nut you need: Python 2

Peter Prettenhofer 119 Oct 08, 2022
Wind Speed Prediction using LSTMs in PyTorch

Implementation of Deep-Forecast using PyTorch Deep Forecast: Deep Learning-based Spatio-Temporal Forecasting Adapted from original implementation Setu

Onur Kaplan 151 Dec 14, 2022
Text Analysis & Topic Extraction on Android App user reviews

AndroidApp_TextAnalysis Hi, there! This is code archive for Text Analysis and Topic Extraction from user_reviews of Android App. Dataset Source : http

Fitrie Ratnasari 1 Feb 14, 2022
Code for evaluating Japanese pretrained models provided by NTT Ltd.

japanese-dialog-transformers 日本語の説明文はこちら This repository provides the information necessary to evaluate the Japanese Transformer Encoder-decoder dialo

NTT Communication Science Laboratories 216 Dec 22, 2022
What are the best Systems? New Perspectives on NLP Benchmarking

What are the best Systems? New Perspectives on NLP Benchmarking In Machine Learning, a benchmark refers to an ensemble of datasets associated with one

Pierre Colombo 12 Nov 03, 2022
Chinese NewsTitle Generation Project by GPT2.带有超级详细注释的中文GPT2新闻标题生成项目。

GPT2-NewsTitle 带有超详细注释的GPT2新闻标题生成项目 UpDate 01.02.2021 从网上收集数据,将清华新闻数据、搜狗新闻数据等新闻数据集,以及开源的一些摘要数据进行整理清洗,构建一个较完善的中文摘要数据集。 数据集清洗时,仅进行了简单地规则清洗。

logCong 785 Dec 29, 2022
jiant is an NLP toolkit

jiant is an NLP toolkit The multitask and transfer learning toolkit for natural language processing research Why should I use jiant? jiant supports mu

ML² AT CILVR 1.5k Jan 04, 2023
NLPShala , the best IDE for all Natural language processing tasks.

The revolutionary IDE for all NLP (Natural language processing) stuffs on the internet.

Abhi 3 Aug 08, 2021
The repository for the paper: Multilingual Translation via Grafting Pre-trained Language Models

Graformer The repository for the paper: Multilingual Translation via Grafting Pre-trained Language Models Graformer (also named BridgeTransformer in t

22 Dec 14, 2022