TURF
Locked
 
|
Hi All,
I will search the archives more completely, but in the TURF bundle, how do I specify a fixed set of items vs. a group of variable items being considered as additions to this fixed set. More specifically, I have products A, B, and C. I want to know which of D, E, or F adds the most incremental reach and frequency to the current fixed A/B/C set. And I have a frequency question (purchases per year) but it's not clear where this "weight" would be entered either. Any help appreciated! Bob Walker www.safllc.com Sent from my iPhone ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
|
|
TURF does not have a "force"
option. If, however, the A,B,C set is near optimal for a triple,
you may be able to get an answer by specifying 4 as the "Maximum Variable
Combinations" and using a large number for "Number of Combinations
to Display". Then scan down the list for the first entry that
include A, B, and C.
As for weights, you can specify importance weights for variables but not for cases. To get the effect of case weights, you would need to replicate cases according to the weights (normalized to a reasonable number so as not to overly inflate the problem) and feed that data into TURF. I have a long list of enhancement requests for TURF, and both of these are on it, but I can't promise any specific delivery date. The most recent addition was scalability to a much larger number of variables than was practical before. Some test problems that previously ran for three days or longer now run in 30 minutes at the cost of a bit of approximation in the answer. I, well, IBM, has a patent application pending for this new approach. The version on the web site is newer than what is shipped with Statistics currently. Jon Peck (no "h") aka Kim Senior Software Engineer, IBM [hidden email] phone: 720-342-5621 From: Bob Walker <[hidden email]> To: [hidden email] Date: 07/28/2015 06:52 PM Subject: [SPSSX-L] TURF Sent by: "SPSSX(r) Discussion" <[hidden email]> Hi All, I will search the archives more completely, but in the TURF bundle, how do I specify a fixed set of items vs. a group of variable items being considered as additions to this fixed set. More specifically, I have products A, B, and C. I want to know which of D, E, or F adds the most incremental reach and frequency to the current fixed A/B/C set. And I have a frequency question (purchases per year) but it's not clear where this "weight" would be entered either. Any help appreciated! Bob Walker www.safllc.com Sent from my iPhone ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
|
|
I had a feeling I might hear from you on this issue – glad I did – thank you! As you surely know, TURF was originally developed for the advertising world to optimize media buying
decisions. But, it is also very widely used in product development and line extension decisions, and especially so to identify new SKUs that best extend reach and frequency off of an existing product line. It is less common to use TURF to build a line of new
products entirely from scratch, which is how the current extension bundle attacks the problem. So, so I am clear, TURF ignores any case weights that may be active? And, would I be correct in assuming that, for each of the six purchase intent scale questions in my particular
study (i.e., the three fixed A/B/C, and the three variable D/E/F), and also applying my frequency question (a “times per year” scale), would require that I replicate each record (with a qualifying purchase intent value, say 4/5 for top two boxes) that number
of times in a new data set? So, for example, if I have a frequency scale value of 26, I would have to replicate the respective A-F record 26 times as duplicate rows in the dataset? Thanks again, Bob W. From: SPSSX(r) Discussion [mailto:[hidden email]]
On Behalf Of Jon K Peck TURF does not have a "force" option. If, however, the A,B,C set is near optimal for a triple, you may be able to get an answer by specifying 4 as the "Maximum Variable Combinations"
and using a large number for "Number of Combinations to Display". Then scan down the list for the first entry that include A, B, and C.
|
|
|
Well, I must be suffering from programming
amnesia. I added support for case weights in February, 2009, so if
you have weights on, the results will take that into account automatically.
And it doesn't even require that the cases be replicated as I described,
so the magnitude of the weight is not an issue.
As for the forcing issue, I'm sure I didn't program that in my sleep. Just using a large value for number to display should give you an answer to that question. There are some possible anomalies, though, that are intrinsic to TURFing. It is possible that adding a new variable, D, E, or F, in your example, would knock out A, B, or C, so the best set with an incremental variable might not include all of the forced variables. You might wind up with A,B,D,E as the incremental solution. That could be superior to A,B,C,D. Which solution would you prefer? It might be valuable to know how close the forced variables are to being the optimal solution, so seeing all the reach results in context might be valuable. If you want to send me a real dataset (with names disguised if you want), I can explore this. One possibility would be to modify the algorithm so that it filters the resulting tables to remove nonqualifying items or to report only items with the maximum number of forced variables in the output. Jon Peck (no "h") aka Kim Senior Software Engineer, IBM [hidden email] phone: 720-342-5621 From: Bob Walker <[hidden email]> To: "[hidden email]" <[hidden email]> Cc: Jon K Peck/Chicago/IBM@IBMUS Date: 07/28/2015 07:57 PM Subject: RE: TURF I had a feeling I might hear from you on this issue – glad I did – thank you! As you surely know, TURF was originally developed for the advertising world to optimize media buying decisions. But, it is also very widely used in product development and line extension decisions, and especially so to identify new SKUs that best extend reach and frequency off of an existing product line. It is less common to use TURF to build a line of new products entirely from scratch, which is how the current extension bundle attacks the problem. So, so I am clear, TURF ignores any case weights that may be active? And, would I be correct in assuming that, for each of the six purchase intent scale questions in my particular study (i.e., the three fixed A/B/C, and the three variable D/E/F), and also applying my frequency question (a “times per year” scale), would require that I replicate each record (with a qualifying purchase intent value, say 4/5 for top two boxes) that number of times in a new data set? So, for example, if I have a frequency scale value of 26, I would have to replicate the respective A-F record 26 times as duplicate rows in the dataset? Thanks again, Bob W. From: SPSSX(r) Discussion [[hidden email]] On Behalf Of Jon K Peck Sent: Tuesday, July 28, 2015 9:26 PM To: [hidden email] Subject: Re: TURF TURF does not have a "force" option. If, however, the A,B,C set is near optimal for a triple, you may be able to get an answer by specifying 4 as the "Maximum Variable Combinations" and using a large number for "Number of Combinations to Display". Then scan down the list for the first entry that include A, B, and C. As for weights, you can specify importance weights for variables but not for cases. To get the effect of case weights, you would need to replicate cases according to the weights (normalized to a reasonable number so as not to overly inflate the problem) and feed that data into TURF. I have a long list of enhancement requests for TURF, and both of these are on it, but I can't promise any specific delivery date. The most recent addition was scalability to a much larger number of variables than was practical before. Some test problems that previously ran for three days or longer now run in 30 minutes at the cost of a bit of approximation in the answer. I, well, IBM, has a patent application pending for this new approach. The version on the web site is newer than what is shipped with Statistics currently. Jon Peck (no "h") aka Kim Senior Software Engineer, IBM peck@... phone: 720-342-5621 From: Bob Walker <rww@...> To: [hidden email] Date: 07/28/2015 06:52 PM Subject: [SPSSX-L] TURF Sent by: "SPSSX(r) Discussion" <[hidden email]> Hi All, I will search the archives more completely, but in the TURF bundle, how do I specify a fixed set of items vs. a group of variable items being considered as additions to this fixed set. More specifically, I have products A, B, and C. I want to know which of D, E, or F adds the most incremental reach and frequency to the current fixed A/B/C set. And I have a frequency question (purchases per year) but it's not clear where this "weight" would be entered either. Any help appreciated! Bob Walker www.safllc.com Sent from my iPhone ===================== To manage your subscription to SPSSX-L, send a message to LISTSERV@...(not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to LISTSERV@...(not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
|
|
In reply to this post by Robert Walker
Glad your memory cells kicked in overnight (mine are dying faster than I can replace them)! Just knowing that I can use case weights is huge. For now, I will simply run all 4-way combinations to get the best ABC + 1. With six options, there are not that many (if my combinatorial memory is right, 6! / 4! * 2!, or 15).
Having the ‘forced alternative’ option would be helpful, just to streamline the output, but the workaround will do for now. Thanks, Bob From: Jon K Peck [mailto:[hidden email]]
Well, I must be suffering from programming amnesia. I added support for case weights in February, 2009, so if you have weights on, the results will take that into account automatically.
And it doesn't even require that the cases be replicated as I described, so the magnitude of the weight is not an issue.
|
|
|
In reply to this post by Jon K Peck
Does SET SEED have any impact on TURF? Thanks, Bob Walker Surveys & Forecasts, LLC From: SPSSX(r) Discussion [mailto:[hidden email]]
On Behalf Of Jon K Peck Well, I must be suffering from programming amnesia. I added support for case weights in February, 2009, so if you have weights on, the results will take that into account automatically.
And it doesn't even require that the cases be replicated as I described, so the magnitude of the weight is not an issue.
|
|
|
No, TURF uses a Python random number generator,
so SET SEED has no effect on that. The Python generator is initialized
in a random way, so you will get a different sequence every time TURF is
run. If there is a need to generate a repeatable sequence, this could
easily be added to TURF.
Jon Peck (no "h") aka Kim Senior Software Engineer, IBM [hidden email] phone: 720-342-5621 From: Bob Walker <[hidden email]> To: "[hidden email]" <[hidden email]> Cc: Jon K Peck/Chicago/IBM@IBMUS Date: 08/28/2015 06:46 AM Subject: RE: TURF Does SET SEED have any impact on TURF? Thanks, Bob Walker Surveys & Forecasts, LLC www.safllc.com From: SPSSX(r) Discussion [[hidden email]] On Behalf Of Jon K Peck Sent: Wednesday, July 29, 2015 9:05 AM To: [hidden email] Subject: Re: TURF Well, I must be suffering from programming amnesia. I added support for case weights in February, 2009, so if you have weights on, the results will take that into account automatically. And it doesn't even require that the cases be replicated as I described, so the magnitude of the weight is not an issue. As for the forcing issue, I'm sure I didn't program that in my sleep. Just using a large value for number to display should give you an answer to that question. There are some possible anomalies, though, that are intrinsic to TURFing. It is possible that adding a new variable, D, E, or F, in your example, would knock out A, B, or C, so the best set with an incremental variable might not include all of the forced variables. You might wind up with A,B,D,E as the incremental solution. That could be superior to A,B,C,D. Which solution would you prefer? It might be valuable to know how close the forced variables are to being the optimal solution, so seeing all the reach results in context might be valuable. If you want to send me a real dataset (with names disguised if you want), I can explore this. One possibility would be to modify the algorithm so that it filters the resulting tables to remove nonqualifying items or to report only items with the maximum number of forced variables in the output. Jon Peck (no "h") aka Kim Senior Software Engineer, IBM peck@... phone: 720-342-5621 From: Bob Walker <rww@...> To: "[hidden email]" <[hidden email]> Cc: Jon K Peck/Chicago/IBM@IBMUS Date: 07/28/2015 07:57 PM Subject: RE: TURF I had a feeling I might hear from you on this issue – glad I did – thank you! As you surely know, TURF was originally developed for the advertising world to optimize media buying decisions. But, it is also very widely used in product development and line extension decisions, and especially so to identify new SKUs that best extend reach and frequency off of an existing product line. It is less common to use TURF to build a line of new products entirely from scratch, which is how the current extension bundle attacks the problem. So, so I am clear, TURF ignores any case weights that may be active? And, would I be correct in assuming that, for each of the six purchase intent scale questions in my particular study (i.e., the three fixed A/B/C, and the three variable D/E/F), and also applying my frequency question (a “times per year” scale), would require that I replicate each record (with a qualifying purchase intent value, say 4/5 for top two boxes) that number of times in a new data set? So, for example, if I have a frequency scale value of 26, I would have to replicate the respective A-F record 26 times as duplicate rows in the dataset? Thanks again, Bob W. From: SPSSX(r) Discussion [[hidden email]] On Behalf Of Jon K Peck Sent: Tuesday, July 28, 2015 9:26 PM To: [hidden email] Subject: Re: TURF TURF does not have a "force" option. If, however, the A,B,C set is near optimal for a triple, you may be able to get an answer by specifying 4 as the "Maximum Variable Combinations" and using a large number for "Number of Combinations to Display". Then scan down the list for the first entry that include A, B, and C. As for weights, you can specify importance weights for variables but not for cases. To get the effect of case weights, you would need to replicate cases according to the weights (normalized to a reasonable number so as not to overly inflate the problem) and feed that data into TURF. I have a long list of enhancement requests for TURF, and both of these are on it, but I can't promise any specific delivery date. The most recent addition was scalability to a much larger number of variables than was practical before. Some test problems that previously ran for three days or longer now run in 30 minutes at the cost of a bit of approximation in the answer. I, well, IBM, has a patent application pending for this new approach. The version on the web site is newer than what is shipped with Statistics currently. Jon Peck (no "h") aka Kim Senior Software Engineer, IBM peck@... phone: 720-342-5621 From: Bob Walker <rww@...> To: [hidden email] Date: 07/28/2015 06:52 PM Subject: [SPSSX-L] TURF Sent by: "SPSSX(r) Discussion" <[hidden email]> Hi All, I will search the archives more completely, but in the TURF bundle, how do I specify a fixed set of items vs. a group of variable items being considered as additions to this fixed set. More specifically, I have products A, B, and C. I want to know which of D, E, or F adds the most incremental reach and frequency to the current fixed A/B/C set. And I have a frequency question (purchases per year) but it's not clear where this "weight" would be entered either. Any help appreciated! Bob Walker www.safllc.com Sent from my iPhone ===================== To manage your subscription to SPSSX-L, send a message to LISTSERV@...(not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to LISTSERV@...(not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to LISTSERV@...(not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
| Free forum by Nabble | Edit this page |