Last edit by: canadiancow
FT members trying to earn a spot in the top 20:
canadiancow - 155k - Trip Log
DrunkCargo (selfloadingcargo) - Triplog
alc (aaalllccc) - 108,250 wings
philelite - ~82k
yvr76 (yvr-flyer) - ~120k
Clan Lindsay
yultraveler (oliviercn)
biglinguist - ~106k
YVRorange - 105k
Bonaventure - 100k~
travelgirl_SE (travelgirl)
supremekai
canadiancow - 155k - Trip Log
DrunkCargo (selfloadingcargo) - Triplog
alc (aaalllccc) - 108,250 wings
philelite - ~82k
yvr76 (yvr-flyer) - ~120k
Clan Lindsay
yultraveler (oliviercn)
biglinguist - ~106k
YVRorange - 105k
Bonaventure - 100k~
travelgirl_SE (travelgirl)
supremekai
Earn Your Wings - Fourth Edition
#166
A FlyerTalk Posting Legend
Original Poster
Join Date: Sep 2012
Location: SFO
Programs: AC SE MM, BA Gold, SQ Silver, Bonvoy Tit LTG, Hyatt Glob, HH Diamond
Posts: 44,356
I don't think it would have been possible to hit 750k in v3, and I don't think it's possible in v4, unless you literally fly all day every day until the end of the contest.
But given that, who cares if it's in there? They're protecting themselves in case they've missed something obvious, and it doesn't negatively affect any of us.
#168
A FlyerTalk Posting Legend
Original Poster
Join Date: Sep 2012
Location: SFO
Programs: AC SE MM, BA Gold, SQ Silver, Bonvoy Tit LTG, Hyatt Glob, HH Diamond
Posts: 44,356
#170
Join Date: Jan 2013
Location: YVR
Programs: AC 100K SE (*G), Marriott P, SPG G, Hilton D
Posts: 611
Shooting for Top 20
Spent last night planning some routes and took some info from cows and others on here. I'm aiming for about 80,000-90,0000 Wings.
I work for IBM and might try write a simple program to find some optimal routes from YVR with Prairies and Scenic Drive badges. I mean I also have access to a Watson computer and CPLEX to run optimization models but I think that might be overkill!
I work for IBM and might try write a simple program to find some optimal routes from YVR with Prairies and Scenic Drive badges. I mean I also have access to a Watson computer and CPLEX to run optimization models but I think that might be overkill!
#171
Join Date: Feb 2011
Location: YYZ
Programs: AC 25K only, he said through tears from the back of the aircraft...
Posts: 563
Spent last night planning some routes and took some info from cows and others on here. I'm aiming for about 80,000-90,0000 Wings.
I work for IBM and might try write a simple program to find some optimal routes from YVR with Prairies and Scenic Drive badges. I mean I also have access to a Watson computer and CPLEX to run optimization models but I think that might be overkill!
I work for IBM and might try write a simple program to find some optimal routes from YVR with Prairies and Scenic Drive badges. I mean I also have access to a Watson computer and CPLEX to run optimization models but I think that might be overkill!
#172
Join Date: Aug 2009
Location: YLW
Programs: AC- SE100 1MM, Hilton Diamond, Marriott Platinum, National Executive, Nexus/GE
Posts: 4,319
Spent last night planning some routes and took some info from cows and others on here. I'm aiming for about 80,000-90,0000 Wings.
I work for IBM and might try write a simple program to find some optimal routes from YVR with Prairies and Scenic Drive badges. I mean I also have access to a Watson computer and CPLEX to run optimization models but I think that might be overkill!
I work for IBM and might try write a simple program to find some optimal routes from YVR with Prairies and Scenic Drive badges. I mean I also have access to a Watson computer and CPLEX to run optimization models but I think that might be overkill!
I may not be as good looking or as smart as most of you, but I am trying to figure this out with my abacus, no fancy computers!
So for early May over a day and half, I got for $1,000 11 flights, for 3,200 wings & four badges plus the Prairie bage? All in Western Canada! I stretched a four flight trip into 11 flights!
#174
Join Date: Feb 2011
Location: YYZ
Programs: AC 25K only, he said through tears from the back of the aircraft...
Posts: 563
Sign me up! I can't afford to compete but am playing with ITA Matrix a lot daydreaming!
#175
Join Date: Jan 2013
Location: YVR
Programs: AC 100K SE (*G), Marriott P, SPG G, Hilton D
Posts: 611
#176
A FlyerTalk Posting Legend
Original Poster
Join Date: Sep 2012
Location: SFO
Programs: AC SE MM, BA Gold, SQ Silver, Bonvoy Tit LTG, Hyatt Glob, HH Diamond
Posts: 44,356
So right now, my code has two "problems":
1. It can't run through every possible itinerary because it simply takes too long. There are various optimizations that can be done here, and it can be parallelized, but:
-In v3, running it with parameters that made it take 5 minutes resulted in pretty much the same itinerary as 60 minutes, so I think the algorithm is pretty good at selecting optimal itineraries.
-There are 17,963 AC-operated flights in the 10 day period I'm looking at. That is a lot of combinations.
2. Cost cost cost. My algorithm optimizes for the maximum wings per hour. This is actually reasonable when you consider the cost of flights versus the cost of my time. However, you run into situations where an itinerary that earns 90% of the wings might be 50% the cost. Pricing out these itineraries is prohibitively slow, and likely expensive. When you consider a 9 day itinerary may need 7-9 tickets, and a query to the ITA QPX API is 3.5 cents, you can see how that might become an issue. Also, right now, my algorithm runs in memory, locally. Add a web service, and that 5 minutes might become 3 weeks. There are "free" ways of doing pricing queries, but it would be slower.
In my mind, (2) is the bigger issue, but it's also much harder to solve. That being said, if you want to factor cost in, then you might need to solve (1) to run through every itinerary, because a $1000 5 day itinerary that earns 50k wings might be better than a $2000 5 day itinerary that earns 80k wings - but that depends on personal circumstances.
1. It can't run through every possible itinerary because it simply takes too long. There are various optimizations that can be done here, and it can be parallelized, but:
-In v3, running it with parameters that made it take 5 minutes resulted in pretty much the same itinerary as 60 minutes, so I think the algorithm is pretty good at selecting optimal itineraries.
-There are 17,963 AC-operated flights in the 10 day period I'm looking at. That is a lot of combinations.
2. Cost cost cost. My algorithm optimizes for the maximum wings per hour. This is actually reasonable when you consider the cost of flights versus the cost of my time. However, you run into situations where an itinerary that earns 90% of the wings might be 50% the cost. Pricing out these itineraries is prohibitively slow, and likely expensive. When you consider a 9 day itinerary may need 7-9 tickets, and a query to the ITA QPX API is 3.5 cents, you can see how that might become an issue. Also, right now, my algorithm runs in memory, locally. Add a web service, and that 5 minutes might become 3 weeks. There are "free" ways of doing pricing queries, but it would be slower.
In my mind, (2) is the bigger issue, but it's also much harder to solve. That being said, if you want to factor cost in, then you might need to solve (1) to run through every itinerary, because a $1000 5 day itinerary that earns 50k wings might be better than a $2000 5 day itinerary that earns 80k wings - but that depends on personal circumstances.
#177
Join Date: Nov 2013
Location: YYC
Programs: AC SE
Posts: 546
Isn't this what AC entices us to do and they even get paid for it...
#179
Join Date: Sep 2014
Location: sqrt(-united states of apologist)
Programs: *$ Green
Posts: 5,403
So right now, my code has two "problems":
1. It can't run through every possible itinerary because it simply takes too long. There are various optimizations that can be done here, and it can be parallelized, but:
-In v3, running it with parameters that made it take 5 minutes resulted in pretty much the same itinerary as 60 minutes, so I think the algorithm is pretty good at selecting optimal itineraries.
-There are 17,963 AC-operated flights in the 10 day period I'm looking at. That is a lot of combinations.
2. Cost cost cost. My algorithm optimizes for the maximum wings per hour. This is actually reasonable when you consider the cost of flights versus the cost of my time. However, you run into situations where an itinerary that earns 90% of the wings might be 50% the cost. Pricing out these itineraries is prohibitively slow, and likely expensive. When you consider a 9 day itinerary may need 7-9 tickets, and a query to the ITA QPX API is 3.5 cents, you can see how that might become an issue. Also, right now, my algorithm runs in memory, locally. Add a web service, and that 5 minutes might become 3 weeks. There are "free" ways of doing pricing queries, but it would be slower.
In my mind, (2) is the bigger issue, but it's also much harder to solve. That being said, if you want to factor cost in, then you might need to solve (1) to run through every itinerary, because a $1000 5 day itinerary that earns 50k wings might be better than a $2000 5 day itinerary that earns 80k wings - but that depends on personal circumstances.
1. It can't run through every possible itinerary because it simply takes too long. There are various optimizations that can be done here, and it can be parallelized, but:
-In v3, running it with parameters that made it take 5 minutes resulted in pretty much the same itinerary as 60 minutes, so I think the algorithm is pretty good at selecting optimal itineraries.
-There are 17,963 AC-operated flights in the 10 day period I'm looking at. That is a lot of combinations.
2. Cost cost cost. My algorithm optimizes for the maximum wings per hour. This is actually reasonable when you consider the cost of flights versus the cost of my time. However, you run into situations where an itinerary that earns 90% of the wings might be 50% the cost. Pricing out these itineraries is prohibitively slow, and likely expensive. When you consider a 9 day itinerary may need 7-9 tickets, and a query to the ITA QPX API is 3.5 cents, you can see how that might become an issue. Also, right now, my algorithm runs in memory, locally. Add a web service, and that 5 minutes might become 3 weeks. There are "free" ways of doing pricing queries, but it would be slower.
In my mind, (2) is the bigger issue, but it's also much harder to solve. That being said, if you want to factor cost in, then you might need to solve (1) to run through every itinerary, because a $1000 5 day itinerary that earns 50k wings might be better than a $2000 5 day itinerary that earns 80k wings - but that depends on personal circumstances.
For #2, if you can scratch out a bunch of useless itinerary, and/or limit to a limit in stops/ticket, I am sure you could come up with a reasonable sample of routes that need pricing. Once you have that, query them once, for each day. Then let the optimizer fetch the price from a saved table so it does query ITA every time it. I may not be understanding your algorithm at all, but what I understand is that during the optimization process, you will query too many times and it will cost too much moula. So if you query once and store the data, you won't query the same route multiple times. If you budget $500, at 3.5c/query, you could price over 14000 routes. How many routes do you need to price? I am sure you could "thin the heard".
#180
A FlyerTalk Posting Legend
Original Poster
Join Date: Sep 2012
Location: SFO
Programs: AC SE MM, BA Gold, SQ Silver, Bonvoy Tit LTG, Hyatt Glob, HH Diamond
Posts: 44,356
For #1, I am sure you have access to a supercomputer somewhere? Or at least a cluster you could ninja for a few hours over night?
For #2, if you can scratch out a bunch of useless itinerary, and/or limit to a limit in stops/ticket, I am sure you could come up with a reasonable sample of routes that need pricing. Once you have that, query them once, for each day. Then let the optimizer fetch the price from a saved table so it does query ITA every time it. I may not be understanding your algorithm at all, but what I understand is that during the optimization process, you will query too many times and it will cost too much moula. So if you query once and store the data, you won't query the same route multiple times. If you budget $500, at 3.5c/query, you could price over 14000 routes. How many routes do you need to price? I am sure you could "thin the heard".
For #2, if you can scratch out a bunch of useless itinerary, and/or limit to a limit in stops/ticket, I am sure you could come up with a reasonable sample of routes that need pricing. Once you have that, query them once, for each day. Then let the optimizer fetch the price from a saved table so it does query ITA every time it. I may not be understanding your algorithm at all, but what I understand is that during the optimization process, you will query too many times and it will cost too much moula. So if you query once and store the data, you won't query the same route multiple times. If you budget $500, at 3.5c/query, you could price over 14000 routes. How many routes do you need to price? I am sure you could "thin the heard".
But I'm worried that spending $500 would only save me $200