Time Constraint Questions

[Byron Knoll]

Hi, I have a couple questions about the time constraints for the contest:

1) "This roughly corresponds to 500'000/T hours real time on a machine with Geekbench5 score of T"

Should this be 50,000 instead of 500'000?

2) phda9v1.8 is currently listed as having a running time of ~23h. Is this based on the 86182 second run time from the large text compression benchmark? If so, that is closer to "~24" hours.

Would it be possible to update this runtime to be based on running phda9v1.8 on the competition test machine? I think that would be very useful for competitors to predict their compressor runtime on the test machine (much more accurately than using the Geekbench5 score). I also think that number would be more intuitive (i.e. it is on the same "scale" as the "<100h" in the row above).

[Matt Mahoney]

Since the new contest is 10x bigger, you have 10x more time.

[Byron Knoll]

On Friday, May 15, 2020 at 7:59:52 PM UTC-7, Matt Mahoney wrote:

Since the new contest is 10x bigger, you have 10x more time.

The test machine is T=500. From the rules, "500'000/T hours" would be 1000 hours instead of 100 hours.

[Matt Mahoney]

Marcus Hutter cleared this up. Geekbench changed their benchmark so you are right. He is going to update the website.

--
You received this message because you are subscribed to the Google Groups "Hutter Prize" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hutter-prize...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/hutter-prize/447b51ea-c9a7-404a-a63b-17c6b83ea74d%40googlegroups.com.

[Byron Knoll]

I tried doing some benchmarking to estimate the runtime of phda9v1.8 on the competition test machine.

From this post, "HP_2017_May" took 29,400 seconds to run on the test machine. On my computer, the same program takes 9,594 seconds to run. phda9v1.8 takes 86,182 seconds on my machine, so I estimate it would take 264,097 seconds on the test machine (73.36 hours).

On Saturday, May 16, 2020 at 9:26:29 AM UTC-7, Matt Mahoney wrote:

Marcus Hutter cleared this up. Geekbench changed their benchmark so you are right. He is going to update the website.

On Sat, May 16, 2020, 12:26 AM Byron Knoll <by...@byronknoll.com> wrote:

On Friday, May 15, 2020 at 7:59:52 PM UTC-7, Matt Mahoney wrote:

Since the new contest is 10x bigger, you have 10x more time.

The test machine is T=500. From the rules, "500'000/T hours" would be 1000 hours instead of 100 hours.

--
You received this message because you are subscribed to the Google Groups "Hutter Prize" group.

To unsubscribe from this group and stop receiving emails from it, send an email to hutter...@googlegroups.com.