Linha Fina

linha-fina is an intent engine that adapts to changing reference data at run time. It adds and removes intents dynamically, without a fixed training pass over the whole intent set.

How it works:

• each intent is an independent SVM that classifies intent vs not-intent
• each intent uses the other intents' data as "not-intent" training data
• only the most relevant negative samples are used for training, ranked by token_set_ratio
• features are one-hot-encoded keyword vectors
• entity extraction from templates uses simplematch
• entity extraction from wordlists optionally uses Aho-Corasick

Limitations:

• long warm-up time (training happens on first inference)
• does not scale well to a large number of intents
• does not take word position into account
• needs retraining whenever a new intent is added

See the docs for concepts, a quickstart, the full API reference, and tuning and troubleshooting guides.

Install

pip install linha-fina

Usage

from linha_fina.engine import IntentEngine

engine = IntentEngine()

intent_1 = ["hello world", "hey there", "hello"]
intent_2 = ["tell me a joke", "say a joke", "make me laugh"]
intent_3 = ["how is the weather", "what's the weather like",
            "what is the weather outside"]
intent_4 = ["my name is {name}", "call me {name}"]
intent_5 = ["change the color to {color}",
            "change light to {color}",
            "set light to {color}}"]

engine.register_intent("hello", intent_1)
engine.register_intent("joke", intent_2)
engine.register_intent("weather", intent_3)
engine.register_intent("introduce", intent_4)
engine.register_intent("change_color", intent_5, entity_samples={"color": ["red", "blue"]})
# you can register more entity examples after intent creation
engine.register_entity("color", ["yellow", "green"], intent_name="color")

for s in ["hello earth",
          "call me Casimiro",
          "my name is Miro",
          "tell me a joke",
          "what is the weather",
          "tell me a joke about the weather",
          "red blue and green are my 3 favorite colors",
          "red is my favorite color",
          "light to blue",
          "change to green",
          "make color green"
          ]:
    print(s)
    print(engine.calc_intent(s))
    # hello earth
    # IntentMatch(name='hello', slots={}, conf=0.9728585833827678)
    # call me Casimiro
    # IntentMatch(name='introduce', slots=[{'name': 'Casimiro'}], conf=0.9236423200801248)
    # my name is Miro
    # IntentMatch(name='introduce', slots=[{'name': 'Miro'}], conf=0.9949282563482034)
    # tell me a joke
    # IntentMatch(name='joke', slots={}, conf=0.9728801357230902)
    # what is the weather
    # IntentMatch(name='weather', slots={}, conf=0.9427588770219081)
    # tell me a joke about the weather
    # IntentMatch(name='joke', slots={}, conf=0.9728801357230902)
    # red blue and green are my 3 favorite colors
    # IntentMatch(name='change_color', slots={'color': 'green'}, conf=0.9648410361539215)
    # red is my favorite color
    # IntentMatch(name='change_color', slots={'color': 'red'}, conf=0.9142220645436988)
    # light to blue
    # IntentMatch(name='change_color', slots={'color': 'blue'}, conf=0.9845866111672085)
    # change to green
    # IntentMatch(name='change_color', slots={'color': 'green'}, conf=0.9598265049335444)
    # make color green
    # IntentMatch(name='change_color', slots={'color': 'green'}, conf=0.9142364945872282)

Related projects

• OpenVoiceOS/ovos-intent-benchmark compares linha-fina against other intent engines, including MycroftAI/padatious, the engine linha-fina is architecturally based on.
• OpenVoiceOS/OVOS-plugin-manager defines the opm.pipeline plugin interface linha-fina implements.

Credits

Originally an experimental research project by TigreGóticoLda, polished and donated to OpenVoiceOS. Its modernization, integration into OpenVoiceOS, and intent benchmarking were funded by the NGI0 Commons Fund.

This project was funded through the NGI0 Commons Fund, a fund established by NLnet with financial support from the European Commission's Next Generation Internet programme, under the aegis of DG Communications Networks, Content and Technology under grant agreement No 101135429.