(Russian) numerals to numbers

A small text-to-number package to transform a line like this

триста тридцать пять с половиной тысяч отборных солдат США и тринадцать целых двадцать одна сотая процента всей экономики мира против полутора русских землекопов-старообрядцев

to a line like this

335500 отборных солдат США и 13.21 процента всей экономики мира против 1.5 русских землекопов-старообрядцев

Based on yargy and natasha.

Locked on natasha 0.10 because of the bloatness of the later versions. May be updated easily to the latest natasha with a small change of providing an Extractor constructor with a pymorphy instance.

super(Rus2Num, self).__init__(NUMBER, pymorphy2.MorphAnalyzer())

Installation

$ pip install rus2num

Usage

from run2num import Rus2Num

r2n = Rus2Num()
text = "Выплаты за второго-третьего ребенка выросли на пятьсот двадцать пять тысячных процента и составили 90 тысяч рублей"
print(r2n(text))
# Выплаты за 2-3 ребенка выросли на 0.525 процента и составили 90000 рублей

Comparison

There are but a few packages from namely

• NVidia's NeMo,
• Oknolaz,
• SergeyShk and its forks from
  • averkij and
  • flockentanz.

NeMo works well but tends to miss many cases I won't have missed (see the comparison table below).

Oknolaz needs to be fed with extracted numbers only and does many mistakes in that case even so bad choice for us.

SergeyShk does either

• replace_groups — тысяча сто to 1100 but сто двести триста to 400 or
• replace — сто двести триста to 100 200 300 but тысяча сто to 1000 100.

It is obvious that addition should be done on decreasing values only so there are some forks to fix it (the overall code is a mess so that I didn't want to do it myself anyway).

averkij and flockentanz work fine both but have some bugs so I took the second one and fixed them. Also I cover cases like с половиной and одна целая две десятых.