Re: [O] table formula help...

[Thierry Banel]

Le 12/12/2014 18:15, Michael Brand a écrit :
> Hi Thierry
>
>
>> Here is how the spreadsheet handles those cases (without modifiers):
>>   |   |   |   | sum |      mean | prod |
>>   |---+---+---+-----+-----------+------|
>>   | 1 | 2 | 3 |   6 |         2 |    6 |
>>   |   | 2 | 3 |   5 | 1.6666667 |    6 |
>>   |   |   | 3 |   3 |         1 |    3 |
>>   |   |   |   |   0 |         0 |    1 | <--- see
>>   #+TBLFM: $4=vsum($1..$3)::$5=vmean($1..$3)::$6=vprod($1..$3)
> - Isn't the above table content from a different TBLFM with a mode
>   string EN for vmean?:
>
>   #+TBLFM: $4=vsum($1..$3)::$5=vmean($1..$3);EN::$6=vprod($1..$3)

Absolutely. I should not write mails so late in the night.

>
> - All columns without mode string (Org >= 8.0):
>
>   |   |   |   | vsum |     vmean | vprod | vmin | vmax |
>   |---+---+---+------+-----------+-------+------+------|
>   | 1 | 2 | 3 |    6 |         2 |     6 |    1 |    3 |
>   |   | 2 | 3 |    5 |       2.5 |     6 |    2 |    3 |
>   |   |   | 3 |    3 |         3 |     3 |    3 |    3 |
>   |   |   |   |    0 | vmean([]) |     1 |  inf | -inf |
>   #+TBLFM: 
> $4=vsum($1..$3)::$5=vmean($1..$3)::$6=vprod($1..$3)::$7=vmin($1..$3)::$8=vmax($1..$3)
>
> - All columns with mode string EN (Org >= 8.0):
>
>   |   |   |   | vsum |     vmean | vprod | vmin | vmax |
>   |---+---+---+------+-----------+-------+------+------|
>   | 1 | 2 | 3 |    6 |         2 |     6 |    1 |    3 |
>   |   | 2 | 3 |    5 | 1.6666667 |     0 |    0 |    3 |
>   |   |   | 3 |    3 |         1 |     0 |    0 |    3 |
>   |   |   |   |    0 |         0 |     0 |    0 |    0 |
>   #+TBLFM: 
> $4=vsum($1..$3);EN::$5=vmean($1..$3);EN::$6=vprod($1..$3);EN::$7=vmin($1..$3);EN::$8=vmax($1..$3);EN
>
>> This is correct. Orgaggregate should behave in a similar way.
>> Fortunately in its latest version it does.
> Ok, I see the similarity in the case for sum of "no input" with which
> I now agree.

And for vmean on zero-length input:
- Spreadsheet without modifiers: vmean([])
- Orgaggregate: Empty
They agree, in this zero-case both return a special value.


>> * Summary
>> Modifiers are lacking in orgaggregate for it to be fully consistent with
>> the spreadsheet. If someone knowns how to add them easily...
> I would try an approach like
>
> #+TBLNAME: test
> | Item | Value |
> |------+-------|
> | a    |       |
> | a    |     2 |
>
> #+BEGIN: aggregate :table test :cols ("Item" "2 * vsum(Value) + 3 *
> vmean(Value); EN")
> | Item | What column header here? How to specify? |
> |------+------------------------------------------|
> | a    |                                        7 |
> #+END
>
> that has a syntax more towards TBLFM with a Calc expression.

I dreamed about such a syntax when designing orgaggregate in the first
place. But I dismissed it as it was going too far in terms of
re-inventing the wheel.

> It would
> not need a mapping of the aggregation function like in
> orgtbl-aggregate-apply-calc-*-function and would go through these
> steps:
>
> 1) Collect list from aggregated input column "Value":
>
>    => '("" "2")
>
> 2) Convert list to Calc vector depending on mode string, see also
>    test-org-table/references/mode-string-EN and
>    test-org-table/org-table-make-reference/mode-string-EN with their
>    siblings:
>
>    (org-table-make-reference '("" "2") t t nil) => "[0,2]"
>
> 3) Detach Calc expression from mode string and replace input header
>    "Value" (possibly several and different input headers per output
>    header) with Calc vector:
>
>    "2 * vsum(Value) + 3 * vmean(Value); EN" =>
>    "2 * vsum([0,2]) + 3 * vmean([0,2])"
>
> 4) Delegate everything else to Calc, just as org-table-eval-formula
>    does:
>
>    (calc-eval "2 * vsum([0,2]) + 3 * vmean([0,2])") => "7"
>
> It is the same that happens already without orgaggregate as
>
> | Value |
> |-------|
> |       |
> |     2 |
> |-------|
> |     7 |
> #+TBLFM: @>$1 = 2 * vsum(@address@hidden) + 3 * vmean(@address@hidden); EN
>
> where the table formula debugger logs:
>
>     Orig:   2 * vsum(@address@hidden) + 3 * vmean(@address@hidden)
>     $xyz->  2 * vsum(@address@hidden) + 3 * vmean(@address@hidden)
>     @r$c->  2 * vsum([0,2]) + 3 * vmean([0,2])
>     $1->    2 * vsum([0,2]) + 3 * vmean([0,2])
>     Result: 7
>
> Mode strings other than "E" and "N" for orgaggregate should then not
> be too far away, see also org-table-eval-formula.
>
> Michael
>

Seems doable.
Would tie the spreadsheet and orgaggregate seamlessly.
Very appealing!

Are you willing to help me implement those steps?

Thierry