Re: [random sorting] (was: org table: one column of random numbers (but natural ones))

[Stig Brautaset]

Uwe Brauer <oub@mat.ucm.es> writes:

>>>> "SB" == Stig Brautaset <stig@brautaset.org> writes:
>
>    >> > In row 67 you would have a random integer in the range [0..67)
>    >> > f0 format removes any fractional part leaving only an integer number
>    >> 
>    >> Aha thanks, a minor thing, which I thank, cannot be really done:
>    >> 
>    >> Is it possible to avoid number repetition?
>    >> 
>    >> So I want a random sequence of the column 67 but I don't want numbers to
>    >> be repeated.
>
>    > To avoid duplicates you could generate a sequence from [0..67), shuffle
>    > it[1], then use the row number as an index into that list. (Or pop off 
> the
>    > front.) How to do that from an org table function I have no idea,
>    > however.
>
> Thanks I tried in a row of 33
> $5=random([0..34]);f0
> $5=random([0..34));f0
> $5=random([0..33));f0
>
> But random repeats, however org-table-sort-lines sorts anyway
> And what I truly needed is a random sorting of sorts.

Right, I think I failed to make myself understood so here's an example
of what I had in mind. It's not convenient to use (need to execute
a source code block) but hopefully what I mean is a bit clearer, and
someone can clean it up a little :-)

First we need to generate a randomised sequence of unique integer in a
range, using the Knuth shuffle I pointed to earlier. Every time you run
tihs you get a new sequence. I've kept the output more to verify that
the results have unique output, as the table formula later will read
from the lisp variable, IIUC.

#+name: random-seq
#+begin_src emacs-lisp :var length=10 :results list
(defun nshuffle (sequence)
  (loop for i from (length sequence) downto 2
        do (rotatef (elt sequence (random i))
                    (elt sequence (1- i))))
  sequence)

(setq random-seq (nshuffle (number-sequence 0 (1- length))))
#+end_src

#+RESULTS: random-seq
- 5
- 9
- 6
- 0
- 7
- 8
- 3
- 2
- 1
- 4

And now for the table that uses the variable. As we compute the
randomised sequence ahead of time for each invocation of the column
formula, and we can use each row number as an index into the sequence to
assign a unique, randomised integer from a range to each column.

| 0 | 5 |
| 1 | 9 |
| 2 | 6 |
| 3 | 0 |
| 4 | 7 |
| 5 | 8 |
| 6 | 3 |
| 7 | 2 |
| 8 | 1 |
| 9 | 4 |
#+TBLFM: $2='(nth (string-to-number $1) random-seq)


Possible improvements (that I don't think I'm up to making):

1. Don't require the column of indices to use as index into the sequence
2. Show how to do this without having the separate pre-compute step of
   the index (possibly with memoizing a sequence on first use?)

Hope this helps!

Regards,

Stig