It's amazing how accurate "out of sight, out of mind" can be, so I better update while I remember.
Over the past week I have been working on my technical report and analysing/collecting more results to affirms some of my reasoning. Unfortunately an electrical fire caused the entire Queen Mary building to be evacuated with all the electrics being turned off (goodbye computer/results). This week I will continue with the my write up of the multi-matching, with the hope of finally moving to the dynamic work ASAP.
As for a plan for the next few weeks, the next Research Degree Meeting is in September and I am in dire need of making my MPhil/PhD transfer (making it 24 months instead of the original 18, or the later 21 months), so in regard to this, I would like all the static graph work complete. Additionally I would like to have implemented the Dynamic graph drawing technique given by Veldhuizen in preparation of my own contribution, with about 30-50% of my thesis written (only in draft).
Now you may be thinking this is quite a steep expectation from myself, me too, but I can try - and if it succeeds, I'll be laughing. If, for whatever reason the plan fails, plan B (Panic) will become my immediate priority.
For now thats all, I shall update again if and when I make any progress. Enjoy the Olympics!
Monday, 30 July 2012
Thursday, 19 July 2012
Survival
It worries me that I wrote an update for this but instead of being published, it was saved, and I didn't notice - even after checking back here several times.
Nonetheless, a brief update of the past two weeks.
Week A represents the week I participated in the iV2012 conference, and overall, I feel it went well. I met with some other members of the Graph Drawing community and discussed some of the problems weve been having and are dealing with at the moment (primarily, designing a structure for dynamic graphs where positions of finer vertices can be used to update coarser vertices in the multilevel structure).
The session itself flopped at the start with no chair turning up. By popular demand (of 2 people) I chaired the session myself (titled Application of Graph Theory) which again went well - I was even thanked afterwards, which although customary, was a confidence boost.
The presentation also went well, with a good response from the audience. Though questions related more to the measuring of aesthetics than the algorithm. No fires, no murders and no horrible hangovers, overall, a good week.
Week B representing this week so far, has so far entailed my return to the technical report which, like everything else I do, has taken forever. Thankfully the conference has given my a little boost to finally finish this piece of work and move on to the bigger problem. Previously, results collected regarding the multi-matching showed a massive increase in runtime with MGF taking almost 4 times as long as the QT - contrary to the single edge contraction implementation.
This has since been fixed and now results are as expected (or not so if the data is correct - which I will explain another time). Unfortunately the edge crossings have now increased to a similar ration (4x) though some results show a decrease to normal levels for matchings of more than 3 vertices. The cause is being investigated.
Week C representing future work: GETTING THIS DONE ASAP! Thats all for now though, see you next time readers.
Nonetheless, a brief update of the past two weeks.
Week A represents the week I participated in the iV2012 conference, and overall, I feel it went well. I met with some other members of the Graph Drawing community and discussed some of the problems weve been having and are dealing with at the moment (primarily, designing a structure for dynamic graphs where positions of finer vertices can be used to update coarser vertices in the multilevel structure).
The session itself flopped at the start with no chair turning up. By popular demand (of 2 people) I chaired the session myself (titled Application of Graph Theory) which again went well - I was even thanked afterwards, which although customary, was a confidence boost.
The presentation also went well, with a good response from the audience. Though questions related more to the measuring of aesthetics than the algorithm. No fires, no murders and no horrible hangovers, overall, a good week.
Week B representing this week so far, has so far entailed my return to the technical report which, like everything else I do, has taken forever. Thankfully the conference has given my a little boost to finally finish this piece of work and move on to the bigger problem. Previously, results collected regarding the multi-matching showed a massive increase in runtime with MGF taking almost 4 times as long as the QT - contrary to the single edge contraction implementation.
This has since been fixed and now results are as expected (or not so if the data is correct - which I will explain another time). Unfortunately the edge crossings have now increased to a similar ration (4x) though some results show a decrease to normal levels for matchings of more than 3 vertices. The cause is being investigated.
Week C representing future work: GETTING THIS DONE ASAP! Thats all for now though, see you next time readers.
Thursday, 5 July 2012
Whoops... Better Update
Every time I come to this I begin writing, then get immediately distracted by something shiny and forget about it.
So a quick catch up, last week I was away from work on holiday - so the gap in my posts can be explained by that (this is more for future me when I come to write up my logbook... btw Carl, get back to work). You'll all be happy to know the cat is fine. Before the holiday, I was setting up and began running the tests for my latest multi-matching implementation - with some analysis of the results thrown in there.
This week has seen a continuation of that analysis in time for a meeting with my supervisors. The results showed that the MGF (CA) tree is much more susceptible to the multi-matching (for the graphs 3elt and data in only 2D layout), showing a large decrease in runtime for matchings of 3 and 4, with a smaller decrease for 5 and leveling out thereafter. The Octree also showed this decrease in runtime, but was not as significant.
An issue with the results is that the runtimes recorded for the MGF tree are much higher than the Octree (contrary to previous implementations). This is almost definitely due to the way it has been implemented (the MGF algorithm was almost completely re-written whereas the OT algorithm remained the same). Other issues show that the results for 3D layouts increase in runtime for the matchings, for BOTH MGF and Octree, suggesting more implementation woes.
Results for 55grid shows some interested spikes in the number of edge crossings for matchings of 3 and 5 vertices (prevalent in both 2D and 3D resultsets), which my supervisors and I believe to be due to the structure of the graph and itself having "preferences" (and sentient thought...).
Since the analysis I have been preparing for the IV2012 conference which is next week. I have been working on my presentation since Yesterday though little progress has been made to change the content since my previous presentations.
Tomorrow will be spent panicking and practising the (hopefully) finished presentation. For now, thats all folks, I shall update next week during or after the conference (if I survive). Au revoir!
So a quick catch up, last week I was away from work on holiday - so the gap in my posts can be explained by that (this is more for future me when I come to write up my logbook... btw Carl, get back to work). You'll all be happy to know the cat is fine. Before the holiday, I was setting up and began running the tests for my latest multi-matching implementation - with some analysis of the results thrown in there.
This week has seen a continuation of that analysis in time for a meeting with my supervisors. The results showed that the MGF (CA) tree is much more susceptible to the multi-matching (for the graphs 3elt and data in only 2D layout), showing a large decrease in runtime for matchings of 3 and 4, with a smaller decrease for 5 and leveling out thereafter. The Octree also showed this decrease in runtime, but was not as significant.
An issue with the results is that the runtimes recorded for the MGF tree are much higher than the Octree (contrary to previous implementations). This is almost definitely due to the way it has been implemented (the MGF algorithm was almost completely re-written whereas the OT algorithm remained the same). Other issues show that the results for 3D layouts increase in runtime for the matchings, for BOTH MGF and Octree, suggesting more implementation woes.
Results for 55grid shows some interested spikes in the number of edge crossings for matchings of 3 and 5 vertices (prevalent in both 2D and 3D resultsets), which my supervisors and I believe to be due to the structure of the graph and itself having "preferences" (and sentient thought...).
Since the analysis I have been preparing for the IV2012 conference which is next week. I have been working on my presentation since Yesterday though little progress has been made to change the content since my previous presentations.
Tomorrow will be spent panicking and practising the (hopefully) finished presentation. For now, thats all folks, I shall update next week during or after the conference (if I survive). Au revoir!
Subscribe to:
Posts (Atom)