iacs computational challenge!. contact info pavlos protopapas [email protected] team lead...

11
IACS Computational Challenge!

Upload: lucy-baldwin

Post on 17-Jan-2016

216 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: IACS Computational Challenge!. Contact info Pavlos Protopapas pprotopapas@cfa.harvard.edu TEAM LEAD Bpprotopapas@cfa.harvard.edu Mauricio Santillana msantill@fas.harvard.edu

IACS Computational Challenge!

Page 2: IACS Computational Challenge!. Contact info Pavlos Protopapas pprotopapas@cfa.harvard.edu TEAM LEAD Bpprotopapas@cfa.harvard.edu Mauricio Santillana msantill@fas.harvard.edu

Contact info• Pavlos Protopapas [email protected] TEAM LEAD B

• Mauricio Santillana [email protected] TEAM LEAD A

• Ozlem Ergun [email protected] PROBLEM COORDINATOR

• Rosalind Reid [email protected] CHALLENGE COORDINATOR

• Robert Parrott [email protected] [email protected] COMPUTATIONAL

• Natasha Baker [email protected] EVERYTHING ELSE

Page 3: IACS Computational Challenge!. Contact info Pavlos Protopapas pprotopapas@cfa.harvard.edu TEAM LEAD Bpprotopapas@cfa.harvard.edu Mauricio Santillana msantill@fas.harvard.edu

Dates• Tuesday the 10th at 10:00am kickoff meeting.

Introductions.Setting up computer accounts etcGetting the data. Explain the rules.

• Friday the 13th at 3:30pm. Preliminary 20 minutes presentation to IACS advisory board.

• Thursday 19th at 1:30m dry run.

• Friday 20th 9:00am final submission of the code and schedule.

– Time limit on running will be 3 hours. If code crashes, teams will have to complete in those 3 hours.

• Friday the 20th at 3:45pm final presentation (S. Wolfram and others will be present)

Page 4: IACS Computational Challenge!. Contact info Pavlos Protopapas pprotopapas@cfa.harvard.edu TEAM LEAD Bpprotopapas@cfa.harvard.edu Mauricio Santillana msantill@fas.harvard.edu

PRICES

• iPADS for the winners• Certificates for all • IACS log FLASH drives for the loosing team.

Page 5: IACS Computational Challenge!. Contact info Pavlos Protopapas pprotopapas@cfa.harvard.edu TEAM LEAD Bpprotopapas@cfa.harvard.edu Mauricio Santillana msantill@fas.harvard.edu

METRIC

• The objective is the sum of unmet demand after every period.

• In case of a tie, best presentation wins

Page 6: IACS Computational Challenge!. Contact info Pavlos Protopapas pprotopapas@cfa.harvard.edu TEAM LEAD Bpprotopapas@cfa.harvard.edu Mauricio Santillana msantill@fas.harvard.edu

Computational resources.

• Each team will be given access to 3 computing nodes with GPGPU.

Page 7: IACS Computational Challenge!. Contact info Pavlos Protopapas pprotopapas@cfa.harvard.edu TEAM LEAD Bpprotopapas@cfa.harvard.edu Mauricio Santillana msantill@fas.harvard.edu

TEAMS• TEAM A: Mauricio Santillana

• Qin Yu

• Ariana Minot

• Yifan Wu

• TEAM B: Pavlos Protopapas

• Chris Beaumont

• Ye Zhao

• Blessing Okeke

Page 8: IACS Computational Challenge!. Contact info Pavlos Protopapas pprotopapas@cfa.harvard.edu TEAM LEAD Bpprotopapas@cfa.harvard.edu Mauricio Santillana msantill@fas.harvard.edu

Budget

• $100 per team for pizza

Page 9: IACS Computational Challenge!. Contact info Pavlos Protopapas pprotopapas@cfa.harvard.edu TEAM LEAD Bpprotopapas@cfa.harvard.edu Mauricio Santillana msantill@fas.harvard.edu

Problem Statement• Attached are two excel files. Each has a list of nodes, each with supply or demand and coordinates. All

supply and demand is present immediately and can be used with no limit in any period. There is also a list of undirected edges, with a weight corresponding to how many resources are required to open that edge. The last column is how much resources each period has to clear debris.

• The objective is the sum of unmet demand after every period.• The constraint is you can start opening arcs in a connected manner from any node with supply, although it

might be easier to say edges can be opened arbitrarily.

Page 10: IACS Computational Challenge!. Contact info Pavlos Protopapas pprotopapas@cfa.harvard.edu TEAM LEAD Bpprotopapas@cfa.harvard.edu Mauricio Santillana msantill@fas.harvard.edu

Problem Statement• gridSample:• 10x10 grid with one source node, all arcs cost 1 to open. Should be pretty ismple to get a good solution.

We have near optimal solutions to this.

• cambridge:

• Two source nodes, one is big and can service the entire network itself, and one is small but can only service part of the network alone. I was going to put capacities on how much each can do a period, but this has already taken a long time to get set up. The data for debris, demand, and supply come from Melih, I had to create how much resources we had to clear debris each period and balance the supply/demand numbers since they aren't scaled.

• Two things of note. First, the resources available to clear debris increase every three periods. The debris can be cleared in 8 periods easy enough, and the model indicates 9 available. Second, Melih provided codes indicating primary vs secondary roads. The assumption I made on the primary roads is that they are wider and should be easier to clear a path, so multiplied the given cost to open primary road arcs by a random number between 0 and 1.

Page 11: IACS Computational Challenge!. Contact info Pavlos Protopapas pprotopapas@cfa.harvard.edu TEAM LEAD Bpprotopapas@cfa.harvard.edu Mauricio Santillana msantill@fas.harvard.edu

Problem Statement• gridSample:

10x10 grid with one source node, all arcs cost 1 to open. Should be pretty ismple to get a good solution. We have near optimal solutions to this.

• cambridge:

Two source nodes, one is big and can service the entire network itself, and one is small but can only service part of the network alone. I was going to put capacities on how much each can do a period, but this has already taken a long time to get set up. The data for debris, demand, and supply come from Melih, I had to create how much resources we had to clear debris each period and balance the supply/demand numbers since they aren't scaled.

Two things of note. First, the resources available to clear debris increase every three periods. The debris can be cleared in 8 periods easy enough, and the model indicates 9 available. Second, Melih provided codes indicating primary vs secondary roads. The assumption I made on the primary roads is that they are wider and should be easier to clear a path, so multiplied the given cost to open primary road arcs by a random number between 0 and 1.