pyfmr: a python wrapper for FMR

A glance overview of what can FMR do

// semantic parsing
"五与5.8的和的平方的1.5次方与two的和减去261.712" =>
nf.math.sub(
  nf.math.sum(
    nf.math.pow(
      nf.math.pow(
        nf.math.sum(
          5,
          nf.math.to_number("5.8")
        ),
        2
      ),
      nf.math.to_number("1.5")
    ),
    2
  ),
  nf.math.to_number("261.712")
); // denotation: 1000

// slot filling
"从上海到天津的机票" => nf.flight("上海", "天津");
"到重庆，明天，从北京" => nf.flight("北京", "重庆");
"到上海去" => nf.flight(null, "上海");

Usage

examples

Install pyfmr via pip

pip install pyfmr

Grammar file

Save the following grammar content to file sf.grammar

<flight> = <departure> <arrival> {nf.flight($1, $2)};
[flight] = <arrival> <departure> {nf.flight($2, $1)};

<departure> = <from> <city> {nf.I($2)};

<arrival> = <to> <city> {nf.I($2)};
[arrival] = <arrival> {nf.arrival($1)};

<from> = "从" ;

<to> = "到" | "去" | "飞";

<city> = "北京"       {nf.I($@)}
       | "天津"       {nf.I($@)}
       | "上海"       {nf.I($@)}
       | "重庆"       {nf.I($@)}
       | `.(?:城|都)` {nf.I($@)}
       ;

<city_ext> = <city>            {nf.I($1)}
           | (any{1,1}) <city> {nf.I($2)}
           ;

<cities> = "直辖市：" (list<city_ext>) {nf.I($@)};

Python example codes

import pyfmr

p = pyfmr.Parser("./sf.grammar")
strs = [
    "直辖市：北京、上海、天津",
    "直辖市：帝都、津城、魔都",
    "**现在有四个直辖市：帝都、魔都、天津、重庆。",
    "**曾经的直辖市：帝都、魔都、天津、重庆、旧都。",
    "天津大学",
]
for l in strs:
    ret = p.extract(l, "cities")
    for item in ret:
        print(item)
    print("="*80)

How to build

The cross compiling is done by xgo.

Prerequisites

Docker
Golang

Steps

# 1. clone the git repo
git clone https://github.com/liuzl/pyfmr && cd pyfmr

# 2. pull xgo docker image
docker pull karalabe/xgo-latest

# 3. install xgo
go get github.com/karalabe/xgo

# 4. build shared libraries
xgo --targets=*/amd64 -buildmode=c-shared -out pyfmr/lib/fmr github.com/liuzl/pyfmr/src

# 5. build wheel
python3 setup.py sdist bdist_wheel

# 6. upload to pypi.org
python3 -m twine upload dist/*