tsort1=:] {&.>~ [: < [: tgrade [
..is also useful. Sorts inverted table y by inverted table x (subset of
columns of y)
,.each key tsort1 invtble
,.each tsort1~ tfreqtble key NB. Equivalent of tsort
,.each (tnub key) tsort1 tfreqtble keyOh and x0 defined in the previous post would be better defined as follows: x0=: 'ID',"1 ];._1 ' ',": 10000+ i.40 --Sherlock, Ric wrote: > I'm investigating working with data stored in inverted tables. > <http://www.jsoftware.com/jwiki/Essays/Inverted_Table> > > Some example data: > hdr=: ;: 'ID DOB Sex Nat Height Weight Arrests' > x0=: 'ID',"1 ];._1 ' ',": 10000+ 40 [EMAIL PROTECTED] 4 > x1=: 1998+ 40 [EMAIL PROTECTED] 4 > x2=: (40 [EMAIL PROTECTED] 2){ >'Female';'Male' > x3=: (40 [EMAIL PROTECTED] 3){ >'NZ';'US';'CH' > x4=: 1.2 + 1 * ([EMAIL PROTECTED] 0) > x5=: 60 + 12 * ([EMAIL PROTECTED] 0) > x6=: 40 [EMAIL PROTECTED] 6 > invtble=: x0;x1;x2;x3;x4;x5;x6 > > I came up with the following to give the frequency of each > unique rows in an inverted table: > tfreq=: #/.~@:|:@:(i.&>)~ > > key=: 1 3 2{invtble > ,.each tfreq key > > The following verb makes a frequency table: > tfreqtble=: [: tsort tnub , <@:tfreq > > ,.each tfreqtble key > > Or a more generalised version > tkeytble=: [: tsort ([: tnub [) , [: boxopen ] > > ,.each key tkeytble tfreq key > > And I can calculate say the sums of data columns by the set > of key columns e.g. > tkeysum=: ] +//.&.>~ <@:tindexof~@:[ > > dat=: 4 5 6{invtble > ,.each key tkeysum dat > > But what I think I really want is a verb tkey, that does what > /. does, but on inverted tables. > Then instead of defining tkeysum, tkeymin, tkeymax ..., I > could do this: > key +/tkey dat > key <./tkey dat > key >./tkey dat > ... > Or in a sorted table: > ,.each key tkeytble key +/tkey dat > > Any pointers or other approaches to consider gratefully accepted! ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
