Hmm... I imagine customers buying multiple TVs would tend to be business owners and at least some of them might be inclined to only return defective TVs.
Anyways, as currently specified, the partial returns issue seems to turn this into the packing problem for the general case. You should expect either exponential growth in resource costs or occasional non-optimal results because of that. Given that issue, why should this be thought of as a time critical algorithm? (Put differently, if there were a transaction number associated with each transaction - and printed on the receipt and the shipping label and so on - this becomes trivial for the typical case, and you have a reasonable conversation which can take place for the unusual cases.) Thanks, -- Raul On Wed, Apr 12, 2017 at 3:16 PM, Joe Bogner <[email protected]> wrote: > Yes, Partial Returns would be the next step. The goal is still useful, > albeit not complete. Imagine the shopkeeper wanting to know how his > business is performing month-by-month. One interesting view would be net of > returns. For example, if the shopkeeper owns a TV store and wants to look > at his total sales in February. He might want to look back at February in > April (restate February) after all his customers have returned their TVs > from the Super Bowl. Certainly there will be some customers who bought 2 > TVs, and only returned 1, but many would return the whole lot. > > All that aside, I'm more interested in how an algorithm such as this could > be implemented in J > > applycredits5 is the best yet -- using amend and only scanning rows prior > to the current return > > NB. 1.813 vs 4.03 from the previous version (j64-806) > (6!:2) 'applycredits5 gen 3e4' > 1.81315 > > Also grows linearly(?) > > 'CANCEL CREDIT OK' =: i.3 > applycredits5 =: 3 : 0 > goals=:(#keys)#OK > credits=. I. (qtys<0) > cancelled=.0 > for_i. credits do. > skeys =. i {. keys > sqtys=. i {. qtys > sseqs =. i {. seqs > > key=.i{keys > seq=.i{seqs > qty=.i{qtys > orig =. {. sseqs #~ > ((sqtys=(_1*qty))*(key=skeys)*(sseqs<seq)*(sqtys>0)*(0=cancelled e.~ sseqs)) > if. orig do. > cancelled=.cancelled,orig > orig_idx=. (seqs i. orig) > goals=: CANCEL orig_idx } goals > goals=: CREDIT i } goals > end. > > end. > goals > ) > > On Wed, Apr 12, 2017 at 2:55 PM, Raul Miller <[email protected]> wrote: > >> This goal concept seems dubious. >> >> Consider, for example, what should happen when someone buys 3 pens and >> then returns one because it was defective. As structured, it seems to >> require that all three be returned. (Presumably a clever shopkeeper >> who understood all the details of how this system works would cancel >> all three and then "sell" two more. But anyone who watches the news >> presumably knows that not everyone is that kind of clever shopkeeper.) >> >> But it's hard to think of sensible code that will work properly once >> the specifications are complete when things are left broken like this. >> >> What should happen when someone tries to cancel a non-zero quantity >> which is less than any available transaction? >> >> Thanks, >> >> -- >> Raul >> >> >> On Wed, Apr 12, 2017 at 12:35 PM, Joe Bogner <[email protected]> wrote: >> > I have a problem I'm trying to solve in different languages. I have a >> > solution in SQL and also in kdb which largely resembles the SQL solution. >> > I'm curious what a J solution would look like. More specifically, I'm >> > interested in picking the brains of others here to see if this type of >> > problem can be solved without looping (some form of scan?). >> > >> > EDIT: Initially I wrote this up thinking the J solution would difficult, >> > but it was actually fairly straightforward -- about 15 minutes, but still >> > would like to see if there are alternatives. If nothing else, maybe an >> > interesting problem to share. >> > >> > Example data: >> > >> > A store has a transaction log with a sequence for each transaction. The >> > transaction log records a key for a unique customer/item combination. The >> > transaction log records how many units were purchased or returned. >> > >> > Goal: >> > Attempt to match up related transactions and cancel out instances when >> the >> > customer/item combination is returned at the same quantity as a previous >> > transaction >> > >> > Examples: >> > >> > Joe buys 1 blue pen, which is defective, then returns the 1 defective >> blue >> > pen, then buys another blue pen. EXPECTED: cancel out first two >> > transactions and leave the the last one for 1 pen >> > >> > Bob buys 2 red pens in two separate transactions. He then buys 3 more. He >> > returns the first two purchases as two separate return transactions. >> > EXPECTED: cancel out all transactions except the one for qty 3 >> > >> > Jane buys 5 purple pens and subsequently returns two of them. She buys >> two >> > more. EXPECTED: No transactions match exactly, so nothing is cancelled >> out >> > >> > >> > Data: >> > >> > data=: 0 : 0 >> > seq key qty >> > 1 1 1 >> > 2 1 _1 >> > 3 1 1 >> > 4 2 1 >> > 5 2 1 >> > 6 2 3 >> > 7 2 _1 >> > 8 2 _1 >> > 9 3 5 >> > 10 3 _2 >> > 11 3 2 >> > ) >> > tbl =: ,. ' ' cut every cutLF data >> > 'seqs keys qtys' =: |: ". every }. tbl >> > >> > >> > Goal: >> > >> > goals =: 0 : 0 >> > >> > goal >> > >> > cancelled >> > >> > credit >> > >> > ok >> > >> > cancelled >> > >> > cancelled >> > >> > ok >> > >> > credit >> > >> > credit >> > >> > ok >> > >> > ok >> > >> > ok >> > >> > ) >> > >> > >> > >> > >> > tbl,.(cutLF goals) >> > >> > +---+---+---+---------+ >> > >> > |seq|key|qty|goal | >> > >> > +---+---+---+---------+ >> > >> > |1 |1 |1 |cancelled| >> > >> > +---+---+---+---------+ >> > >> > |2 |1 |_1 |credit | >> > >> > +---+---+---+---------+ >> > >> > |3 |1 |1 |ok | >> > >> > +---+---+---+---------+ >> > >> > |4 |2 |1 |cancelled| >> > >> > +---+---+---+---------+ >> > >> > |5 |2 |1 |cancelled| >> > >> > +---+---+---+---------+ >> > >> > |6 |2 |3 |ok | >> > >> > +---+---+---+---------+ >> > >> > |7 |2 |_1 |credit | >> > >> > +---+---+---+---------+ >> > >> > |8 |2 |_1 |credit | >> > >> > +---+---+---+---------+ >> > >> > |9 |3 |5 |ok | >> > >> > +---+---+---+---------+ >> > >> > |10 |3 |_2 |ok | >> > >> > +---+---+---+---------+ >> > >> > |11 |3 |2 |ok | >> > >> > +---+---+---+---------+ >> > >> > >> > >> > One approach: >> > >> > applycredits =: 3 : 0 >> > >> > goals=.(<'goal') >> > >> > creditids=.0 >> > >> > for_i. (i. # seqs) do. >> > >> > key=.i{keys >> > >> > seq=.i{seqs >> > >> > qty=.i{qtys >> > >> > nextcredit =.| {. qtys #~ ((key=keys)*(seqs>seq)*(qtys<0)) >> > >> > if. nextcredit = qty do. >> > >> > goals=.goals,<'cancelled' >> > >> > creditids =. creditids, seqs #~ ((key=keys)*(seqs>seq)*(qtys<0)) >> > >> > elseif. creditids e.~ seq do. >> > >> > goals=.goals,<'credit' >> > >> > elseif. do. >> > >> > goals=.goals,<'ok' >> > >> > end. >> > >> > end. >> > >> > goals >> > >> > ) >> > >> > tbl ,. ( applycredits 0 ) >> > >> > >> > +---+---+---+---------+ >> > >> > |seq|key|qty|goal | >> > >> > +---+---+---+---------+ >> > >> > |1 |1 |1 |cancelled| >> > >> > +---+---+---+---------+ >> > >> > |2 |1 |_1 |credit | >> > >> > +---+---+---+---------+ >> > >> > |3 |1 |1 |ok | >> > >> > +---+---+---+---------+ >> > >> > |4 |2 |1 |cancelled| >> > >> > +---+---+---+---------+ >> > >> > |5 |2 |1 |cancelled| >> > >> > +---+---+---+---------+ >> > >> > |6 |2 |3 |ok | >> > >> > +---+---+---+---------+ >> > >> > |7 |2 |_1 |credit | >> > >> > +---+---+---+---------+ >> > >> > |8 |2 |_1 |credit | >> > >> > +---+---+---+---------+ >> > >> > |9 |3 |5 |ok | >> > >> > +---+---+---+---------+ >> > >> > |10 |3 |_2 |ok | >> > >> > +---+---+---+---------+ >> > >> > |11 |3 |2 |ok | >> > >> > +---+---+---+---------+ >> > >> > >> > >> > (cutLF goals) -: ( applycredits 0 ) >> > >> > 1 >> > >> > >> > thanks for any input >> > ---------------------------------------------------------------------- >> > For information about J forums see http://www.jsoftware.com/forums.htm >> ---------------------------------------------------------------------- >> For information about J forums see http://www.jsoftware.com/forums.htm > ---------------------------------------------------------------------- > For information about J forums see http://www.jsoftware.com/forums.htm ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
