Hacker News Reader

Comment by CraigJPerry | Hacker News Reader

A chance to use my favourite feature of excel, flash fill:

Given:

    A       B
    a1b2c3  123
    2g34
    34f5
    4l5p6
    1.23E+06

Selecting column B with the one example cell and hitting flash fill yields:

    A         B
    a1b2c3    123
    2g34      234
    34f5      345
    4l5p6     456
    1.23E+06  12306

I'd still absolutely love to see the code behind this feature.

EDIT: i just noticed an error in the last row, should be 1230000 not 12306. Ahh well.

View on HN

> "I'd still absolutely love to see the code behind this feature."

Windows PowerShell has a cmdlet ConvertFrom-String which "supports automatically-generated, example-driven parsing based on the FlashExtract, research work by Microsoft Research."[1]. Then when they open-sourced Powershell, that was one of the cmdlets which did not come over and is no longer supported, so no way to see the source code there, sadly.

(Video [2] shows that FlashFill and FlashExtract and Excel / PowerShell uses are related, and he says it generates many possible programs which fit the given examples, and uses 'machine learning based ranking techniques' to choose which one fits best).

[1] https://docs.microsoft.com/en-us/powershell/module/microsoft...

[2] https://www.youtube.com/watch?v=w-k9WjRJvIY

I think this is the paper https://arxiv.org/abs/1204.6079

Ah yes, Flash fill. It always falls at the last hurdle. Nice idea, never quite gets it right.

It looks something like:

try { return parse_float(text) } catch ParseError { return eval(match(/\d+/g, text).matches.join("")) }

Unrelated to the implementation: in what circumstances is flash fill useful?

Switch to R1C1 Notation. It makes a lot of Excel magic easier to understand.

Reply on news.ycombinator.com