PyDuckling

Copyright © 2020– Treble.ai

Overview

This package provides native bindings for Facebook's Duckling in Python. This package supports all dimensions and languages available on the original library, and it does not require to spawn a Haskell server and does not use HTTP to call the Duckling API.

Note: This package is completely Haskell-less

Installing

To install pyduckling, you can use both conda and pip package managers:

# Using pip
pip install pyduckling-native

# Using conda
conda install pyduckling -c treble-ai

Notes: Right now, we only provide package distributions for Linux (x86_64). We will provide Windows and Mac distributions on the next release

Dependencies

To compile pyduckling, you will require the latest nightly release of Rust, alongside Cargo. Also, it requires a Python distribution with its corresponding development headers. Finally, this project depends on the following Cargo crates:

• PyO3: Library used to produce Python bindings from Rust code.
• Maturin: Build system to build and publish Rust-based Python packages

Additionally, this package depends on Duckling-FFI, used to compile the native interface to Duckling on Haskell. In order to compile Duckling-FFI, you will require the Stack Haskell manager.

Installing locally

Besides Rust and Stack, you will require the latest version of maturin installed to compile this project locally:

pip install maturin toml

First, you will need to compile Duckling-FFI in order to produce the shared library libducklingffi, to do so, you can use the git submodule found at the root of this repository:

cd duckling-ffi
stack build

Then, you will need to move the resulting binary libducklingffi.so to the ext_lib folder:

cp duckling-ffi/libducklingffi.so ext_lib

After completing this procedure, it is possible to execute the following command to compile pyduckling:

maturin develop

In order to produce wheels, maturin build can be used instead. This project supports PEP517, thus pip can be used to install this package as well:

pip install -U .

Running tests

We use pytest to run tests as it follows (after calling maturin develop):

pytest -v duckling/tests

Package usage

PyDuckling provides access to the parsing capabilities of Duckling used to extract structured data from text.

# Core imports
from duckling import (load_time_zones, parse_ref_time,
                      parse_lang, default_locale_lang, parse_locale,
                      parse_dimensions, parse, Context)

# Load reference time for time parsing
time_zones = load_time_zones("/usr/share/zoneinfo")
bog_now = pendulum.now('America/Bogota').replace(microsecond=0)
ref_time = parse_ref_time(
    time_zones, 'America/Bogota', bog_now.int_timestamp)

# Load language/locale information
lang_es = parse_lang('ES')
default_locale = default_locale_lang(lang_es)
locale = parse_locale('ES_CO', default_locale)

# Create parsing context with time and language information
context = Context(ref_time, locale)

# Define dimensions to look-up for
valid_dimensions = ["amount-of-money", "credit-card-number", "distance",
                    "duration", "email", "number", "ordinal",
                    "phone-number", "quantity", "temperature",
                    "time", "time-grain", "url", "volume"]

# Parse dimensions to use
output_dims = parse_dimensions(valid_dimensions)

# Parse a phrase
result = parse('En dos semanas', context, dims, False)

This wrapper allows access to all the dimensions and languages available on Duckling:

Changelog

Please see our CHANGELOG file to learn more about our new features and improvements.

Contribution guidelines

We follow PEP8 and PEP257 for pure python packages and Rust to compile extensions. We use MyPy type annotations for all functions and classes declared on this package. Feel free to send a PR or create an issue if you have any problem/question.