Padatious

Padatious is a neural network intent parser, implemented in pure numpy with a FANN-compatible model format. This repository packages it as an OpenVoiceOS (OVOS) pipeline plugin and bundles a maintained fork of the original padatious from Mycroft AI.

Status

This plugin is legacy and maintained in a limited capacity; it is not recommended for new deployments. The underlying engine memorizes its training sentences instead of generalizing from them, so it needs many worded variations of each intent to catch paraphrases a classifier would generalize to on its own. Compile time and memory grow with the number of training sentences, and the engine becomes unusable once it is asked to train on the full OVOS skill corpus rather than a handful of intents.

For new deployments, use ovos-m2v-pipeline, a classifier-based intent engine that generalizes from training sentences, or nebulento, a fuzzy-matching intent engine, instead.

Features

• Intents are easy to create from a handful of example sentences.
• Each intent trains its own small network, independent of the others.
• Fast training on a small amount of data.
• Entity extraction from a matched sentence (for example, Find the nearest {place} matches "Find the nearest gas station" and extracts place: gas station).

Installing

Padatious is pure Python (numpy). It needs no native libraries or compilers.

Install from PyPI:

pip install ovos-padatious-pipeline-plugin

Direct Usage

from ovos_padatious import IntentContainer

container = IntentContainer('intent_cache')
container.add_intent('hello', ['Hi there!', 'Hello.'])
container.add_intent('goodbye', ['See you!', 'Goodbye!'])
container.add_intent('search', ['Search for {query} (using|on) {engine}.'])
container.train()

print(container.calc_intent('Hello there!'))
print(container.calc_intent('Search for cats on CatTube.'))

container.remove_intent('goodbye')

Inside OVOS, the plugin is discovered automatically through its opm.pipeline entry point. See docs/ for installation details, the intent file syntax, the full Python API, pipeline configuration, and the matching algorithm.

Inside OVOS, training and compiling always run on a background worker, never on the thread that registered or queried something (including the very first pass); a test or tool that registers an intent and needs to query it right away should call PadatiousPipeline.wait_until_trained() (see docs/ovos_pipeline.md) rather than polling or sleeping.

Related projects

• OpenVoiceOS/ovos-spec-tools: the reference implementation of the OVOS architecture specifications, used here for sentence-template expansion and language tag handling.
• OpenVoiceOS/architecture: the OpenVoiceOS architecture specifications.
• OpenVoiceOS/ovos-plugin-manager: the plugin and entry-point system that loads this pipeline into OVOS.

License

Licensed under the Apache 2.0 license.