Kaggle: Reconstruct Tracks From 75 Gb Of Indicate Data

Fartel Engelbert has told me that at that spot is a novel CERN-sponsored car learning rival at Kaggle.com:

TrackML Particle Tracking Challenge

To brand the even short, the data you lot volition convey to download include v times xv GB develop files summation 1 GB develop sample in addition to 1 GB examination file. H5N1 sample submission has xxx MB, detectors.zip convey 175 kB.




Well, readers whose infrastructure is similar to mine convey already given up. I don't know what to create with 75 GB. On Windows, there's no problem to shop this much information but I would convey to manipulate it with Mathematica in addition to that would clearly live also tedious with 75 GB.




On the other hand, I could run a VirtualBox with unopen to Linux, similar during the Higgs Kaggle contest, but hence I would convey to report whether I convey to allocate unopen to extra difficult disk for the faux Linux difficult disk in addition to appear upward similar problems that I am non experienced with. I exactly don't desire to create that – this dataset is but also big for me.

If such things aren't problem for you, you lot should try. In the begin stage of the rival – 3 months are left – you lot involve to blueprint the nearly accurate algorithm to reconstruct particles' tracks from the points that the huge datasets are composed of.

There volition live unopen to other purpose of the rival that starts inward the summertime where the speed of the calculation volition matter.

The leaderboard shows the begin contestant alongside 222 to convey the marking of exactly 0.46 – hence I believe that there's a lot of room for improvement. The preliminary leaderboard is based on unopen to 29% of the data, the lastly ane volition live based on the remaining 71% of the information hence it may live different.

Most importantly, the prizes are $12k, $8k, $5k (in total, $25k) for the first, second, 3rd place.

Good luck.