Darkness 2011-04-13 12:00:00 UTC
 
Twilight 2011-04-14 12:00:00 UTC
 
Daylight 2011-04-15 12:00:00 UTC
 
Finish 2011-04-20 12:00:00 UTC

Crossword - Comments

Search Comments

Go

Date Submission Comment
2011-04-21 20:19:23 UTC Faster - no ram_li_czech_2
by Ram-Li-Czech
Just noticed this entry. Nice results! I'm lucky it wasn't any faster...
by Nicholas Howe
2011-04-21 17:19:19 UTC lasttry01
by Alfonso Nieto-Castanon
Nice entry!
It would be great most rules (like the scoring function) were properly documented in http://www.mathworks.com/matlabcentral/contest/contests/32/rules
...
by Franck Dernoncourt
2011-04-21 08:55:20 UTC Rapid Weight Loss May Be Harmful
by Nicholas Howe
Congrats! And I second what Abhisek said.
by James White
2011-04-21 08:53:25 UTC let's see 3
by David Smith
Nice variation!

by James White
2011-04-21 08:20:08 UTC lasttry01
by Alfonso Nieto-Castanon
impressively fast though!
by Oli
2011-04-21 07:42:12 UTC Rapid Weight Loss May Be Harmful
by Nicholas Howe
Congrats Nick! Personally I was impressed by your algorithm to pack tightly and often completely (score=0) when min(wordlength)==gridsize.
by Abhisek Ukil
2011-04-21 01:25:36 UTC lasttry01
by Alfonso Nieto-Castanon
The scoring algorithms incorporates cyc and nodes. I posted it in haiku form on the blog, but here it is in normal form: score = k1*result + k2*e(k3*runtime) + k4*max(complexity-10,0) + k5*nodes

Where:
k1 = 0.0025
k2 = 5/900 (0.000555)
k3 = 1/9 (0.111111)
k4 = 1
k5 = 1/900 (0.0011111)
by Alan Chalker
2011-04-20 23:12:58 UTC Rapid Weight Loss May Be Harmful
by Nicholas Howe
Oli -
The title does relate to the algorithm, sort of. I created this one by removing some of the solver calls that were in Valle, hoping to make it faster. I only eliminated ones that were never used on the sample testsuite, but I didn't know for sure whether they might be necessary on the real testsuite. Thus I was concerned that my 'slimming' of the program might be harmful to its score.
by Nicholas Howe
2011-04-20 22:41:21 UTC Rapid Weight Loss May Be Harmful
by Nicholas Howe
I guess it is only fair that you win as your solver was still the best in most cases, I think. Being its founder you were most able to tune it I guess :) (I have no idea myself of how it works)
by Oli
2011-04-20 22:31:44 UTC lasttry01
by Alfonso Nieto-Castanon
you are right, cyc=100, should have fix that! great contest to all and congrats to Nick!
by Alfonso Nieto-Castanon
2011-04-20 21:59:37 UTC Rapid Weight Loss May Be Harmful
by Nicholas Howe
Congratulations :)
Does the title relate to the algorithm somehow?
by Oli
2011-04-20 21:55:35 UTC lasttry01
by Alfonso Nieto-Castanon
Resolution is 15th
by Oli
2011-04-20 21:55:19 UTC lasttry01
by Alfonso Nieto-Castanon
If you compare with a random entry like 'Resolution' by Volkan, yours is half the cpu time and 30,000 less. However it has 100 cycles (10 for volkan), and more nodes, maybe that compensated the good scores?
by Oli
2011-04-20 21:06:22 UTC lasttry01
by Alfonso Nieto-Castanon
something weird with the scoring mechanism... not good enough to contest the winner entry anyway
by Alfonso Nieto-Castanon
2011-04-20 20:58:28 UTC Valles Marinaris
by Nicholas Howe
Misspelled the name of Valles Marineris. How embarrassing! Good thing this one didn't win.
by Nicholas Howe
2011-04-20 20:49:15 UTC The Kitchen Sink
by Nicholas Howe
damn, I forgot to update the board-flipping several times too... too bad!
by Oli
2011-04-20 20:29:08 UTC The Kitchen Sink
by Nicholas Howe
Still waiting to get my hands on this code :) Phew.. that's a long queue even with queue-in-darkness introduction.
by Amitabh Verma
2011-04-20 18:46:02 UTC Faster - no ram_li_czech_2
by Ram-Li-Czech
One more comment:

This solution has the following performance on the 100 provided training scenarios (on my PC):

results: 5940021.00
time: 41.68

by Ram-Li-Czech
2011-04-20 18:38:27 UTC Faster - no ram_li_czech_2
by Ram-Li-Czech
Makes me feel good that I have the lowest score (not all the final submissions in the queue went through, though).

Problem: slow, high node

Three new solvers:

ram_li_czech: modification of james_1, forming 2x3 blocks rather than 2x2

ram_li_czech_3 : pretty proud of this one, it uses only the grid (1:2:end,1:2:end) for placement, zero bogus, nice patterns! Often creates nonsense though :)

ram_li_czech_2: My favourite. Starts placing 2x3 blocks ala ram_li_czech. Once exhausted, starts placing from the most important pieces. No bogus, and the result looks pretty dense. Give it a try! Too bad it is slow...

Have a nice day!


by Ram-Li-Czech
2011-04-20 18:17:42 UTC The Kitchen Sink
by Nicholas Howe
That's a pity!
Would it beat valles marineris?
I see the latter is first but hasn't crossed the 6M
by Oli
2011-04-20 18:04:03 UTC C'est nul! 2
by Claire
Pas si nul que ├ža ;)
by Oli
2011-04-20 17:48:19 UTC The Kitchen Sink
by Nicholas Howe
To fix the bug, change the 9 on line 75 to a 12. With that alteration, this would have been my best entry on the sample testsuite.
by Nicholas Howe
2011-04-20 17:31:09 UTC NBW Combo Yot Lat 2
by Nicholas Howe
This is a combination of Albert's NBW entry and my own Lattice code line, which I hadn't submitted because I overlooked the deadline. May be a lucky break for me, if this code helps me with the overall prize. Fingers crossed...
by Nicholas Howe
2011-04-20 17:25:19 UTC Valles Marinaris
by Nicholas Howe
I think this is my best entry score-wise. It breaks 6M on the sample test suite, barely. We'll see how it does when the queue clears!
by Nicholas Howe
2011-04-20 17:23:37 UTC The Kitchen Sink
by Nicholas Howe
This one has a bug in it, found after the deadline.

When fixed it does break 6M on the sample testsuite, but probably not the real on.
by Nicholas Howe