Two minor variations on Raul's approach avoid the
gerund and the second allows choice of the
key column.
(Apologies for my gratuitous line-throws, as usual)
The kernel of the approach is the same as Raul's:
({."1 </.]) raw NB. partition by column 0
+---------+---------+
|+--+--+-+|+--+--+-+|
||t1|f1|5|||t2|f7|7||
|+--+--+-+|+--+--+-+|
||t1|f2|5|||t2|f8|7||
|+--+--+-+|+--+--+-+|
||t1|f3|5|| |
|+--+--+-+| |
+---------+---------+
Reduce each sub-table to unique elements:
({."1 <@~."1 @|:/.]) raw
+----+----------+---+
|+--+|+--+--+--+|+-+|
||t1|||f1|f2|f3|||5||
|+--+|+--+--+--+|+-+|
+----+----------+---+
|+--+|+--+--+ |+-+|
||t2|||f7|f8| ||7||
|+--+|+--+--+ |+-+|
+----+----------+---+
This is similar to your specification, except
that all elements are boxed.
The following form might be more useful:
]S:0 each ({."1 <@~."1 @|:/.]) raw
+--+--+-+
|t1|f1|5|
| |f2| |
| |f3| |
+--+--+-+
|t2|f7|7|
| |f8| |
+--+--+-+
A small change allows you to choose your key:
0 ( {"1 <@~."1@|:/. ]) raw NB. with column 0
+----+----------+---+
|+--+|+--+--+--+|+-+|
||t1|||f1|f2|f3|||5||
|+--+|+--+--+--+|+-+|
+----+----------+---+
|+--+|+--+--+ |+-+|
||t2|||f7|f8| ||7||
|+--+|+--+--+ |+-+|
+----+----------+---+
An index may also be used with Raul's method:
0({"1 {.`<`{.&spread@|:/. ])raw
+--+----------+-+
|t1|+--+--+--+|5|
| ||f1|f2|f3|| |
| |+--+--+--+| |
+--+----------+-+
|t2|+--+--+ |7|
| ||f7|f8| | |
| |+--+--+ | |
+--+----------+-+
Again. we can simplify things up a bit:
]S:0 each 0 ( {"1 <@~."1@|:/. ]) raw
+--+--+-+
|t1|f1|5|
| |f2| |
| |f3| |
+--+--+-+
|t2|f7|7|
| |f8| |
+--+--+-+
]S:0 each 1({"1 <@~."1@|:/.])raw NB.column 1 !!!
+--+--+-+
|t1|f1|5|
+--+--+-+
|t1|f2|5|
+--+--+-+
|t1|f3|5|
+--+--+-+
|t2|f7|7|
+--+--+-+
|t2|f8|7|
+--+--+-+
NB. All this loses any association between elements
in columns 1 & 2. eg if
rawa=:('t1 t1 t1 t2 t2',.&;:'f1 f2 f3 f7 f8'),.]&.>5 5 6 7 7
]S:0 each ({."1 <@~."1 @|:/.]) rawa
+--+--+---+
|t1|f1|5 6|
| |f2| |
| |f3| |
+--+--+---+
|t2|f7|7 |
| |f8| |
+--+--+---+
... so you've lost the association of 5 with f1 & f2 but not f3,
and 6 with f3 but not f1 or f2. Your opening remarks suggest
this isn't a problem, but it could be for more general data.
I haven't explored timings; the transpose (|:) might be expensive
for large sub-tables.
Mike
Raul Miller wrote:
On 6/8/07, Terrence Brannon <[EMAIL PROTECTED]> wrote:
't1' 'f1' 5
't1' 'f2' 5
't1' 'f3' 5
't2' 'f7' 7
't2' 'f8' 7
...
Let's say that you want the compression of this data to be searchable
by the
first field, hence you want a list in which each item is 3 boxes like
so:
't1' ; (<('f1';'f2';'f3')) ; 5
't2' ; (<('f7';'f8')) ; 7
You have not specified the original format of the data, but let's
say it's a boxed two dimensional array, with as many columns
as sql columns and as many rows as sql rows.
raw=:('t1 t1 t1 t2 t2',.&;:'f1 f2 f3 f7 f8'),.]&.>5 5 5 7 7
First off, observe that you can group things the way you want using a
/. dyad:
({."1 </. ])raw
Now we just need a verb to rearrange things the way you want.
spread=:4 :'x`:6 y'"0 _1
({."1 {.`<`{.&spread@|:/. ])raw
+--+----------+-+
|t1|+--+--+--+|5|
| ||f1|f2|f3|| |
| |+--+--+--+| |
+--+----------+-+
|t2|+--+--+ |7|
| ||f7|f8| | |
| |+--+--+ | |
+--+----------+-+
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
