Using functions to convert slot words into different forms

[guanqun-yang]

I have some templates that look like the following

template = '{CITY1} is {MILE} miles from {CITY2} and {NumLessThan(MILE)} miles from {CITY3}.'

where

• CITY, MILE both comes from lexicons added using editor.add_lexicon().
• NumLessThan() is a function that generates valid words given MILE

  def NumLessThanInt(num1):
  return str(random.choice([x for x in range(6) if x < int(num1)]))

  I know the checklist library could handle slots like CITY1, MILE, CITY2 and CITY3. But is there a way to also support slots like NumLessThan(MILE)?

Any input is appreciated. Thanks in advance!

[marcotcr]

We don't support running a function that depends on the instantiation of the placeholders, but you can achieve the same effect by precomputing pairs or tuples with your function, and using that in the template. To use your example:

import random
import numpy as np
import munch
def NumLessThanInt(num1):
    return str(random.choice([x for x in range(6) if x < int(num1)]))
miles = [1, 2, 3, 4, 5]
# sample 1000 pairs
miles_pair = [munch.Munch({'orig': x, 'less': NumLessThanInt(x)}) for x in np.random.choice(miles, 1000)]
x = editor.template('{city1} is {mile.orig} miles from {city2} and {mile.less} miles from {city3}.', mile=miles_pair, nsamples=100)
x.data[:4]

['Tampa is 4 miles from San Juan and 0 miles from Riverside.', 'Virginia Beach is 5 miles from St. Petersburg and 2 miles from Denver.', 'Corpus Christi is 4 miles from Saint Paul and 1 miles from Columbus.', 'Buffalo is 5 miles from Laredo and 0 miles from San Antonio.']

I put it in a much just because I think it's nice to call mile.orig, but it can also be a list or tuple:

import random
import numpy as np
import munch
def NumLessThanInt(num1):
    return str(random.choice([x for x in range(6) if x < int(num1)]))
miles = [1, 2, 3, 4, 5]
miles_pair = [[str(x), NumLessThanInt(x)] for x in np.random.choice(miles, 1000)]
x = editor.template('{city1} is {mile[0]} miles from {city2} and {mile[1]} miles from {city3}.', mile=miles_pair, nsamples=100)
x.data[:4]

['Fort Worth is 4 miles from New York City and 2 miles from Richmond.', 'Atlanta is 3 miles from Glendale and 2 miles from Saint Paul.', 'Orlando is 1 miles from Santa Ana and 0 miles from Lubbock.', 'Glendale is 1 miles from Glendale and 0 miles from Tampa.']