Pie Analysis Blog - 100 man back to back analysis

[Shane_NC]

It has been brought to my attention that I do not possess a deep knowledge of pie factory theory and execute the pie factory like a "beginner".

http://www.twitch.tv/shane_nc/b/456643042

I decided to objectively do a 100 man analysis via save state in mame, to give me an idea of my averages and the average probability of completing a PF on any given L=5 or higher stage.

( I am familiar with statistics and realize due to a small sample size, like the one used in this blog, these numbers aren't as accurate as they could be say if I did a 1000 or 10K man analysis. This is what I had time to do tonight)

I completed 93/100 Pie Factories.

First 50
-------------
44-6

Second 50
-------------
49-1

In this instance I have roughly a 93% chance to complete any given Pie factory screen.


Here is the complete stream uncut all 100 attempts: no practice, this was turn the stream on and go.

http://www.twitch.tv/shane_nc/b/456643042

========================
I am formally offering a $50 bounty to Kyou-Kun (Christian) if at 9pm EST tomorrow night, you fire up mame save state on level=05 Pie factory and can do better than my first attempt streamed live on twitch of 93/100 PF completed.

http://www.twitch.tv/shane_nc/b/456643042

If you complete 94 or more out of 100 you have won yourself $50, I will even send the $50 to a middle man if you do not want me to have your address. Or I will donate the $50 to any players KO3 airfare that you choose. Challenge is there, put up or shut up.
========================

[corey.chambers]

Nice Shane. A few days before my killscreen game I did the 100 pie factory challenge and only got 89%. You are right there, buddy.

[hchien]

Good job!  92-93% is about what you need to have a very good shot at the killscreen.  With that percentage, I'm surprised you haven't killscreened yet.  You should do the same experiment with the rivets and aim for 92-93% as well.

I'd also recommend doing the save state from the end of the board before (especially with the rivets).  The reason being the random number seed will always start out the same if you save state from the intermission screen or the screen itself.  On the rivets, this has the effect of making the fireballs always appear on the same girder levels assuming the same input (usually running to the left on rivets).  As soon as your input deviates/varies (climb up the ladder at a variable frame), the random number seed with "randomize" but by then 1 or 2 fireballs will have already appeared.

[Jeffw]

Yes, I agree with hank about starting from the previous board. A good idea would be to jump at least one barrel on the previous board then climb the ladder. I once was practicing hammerless rivets from a savestate and doing very well and thought that I had a really good success rate but in actual games I wasn't doing as well, and I think the reason was because I was starting from the same random seed every time with the savestate. Really, the best thing to do if you want to accurately know your success rate would be to use pie factories from actual games, although it's harder to get a large number of them this way.

[Milehighdt]

When I use save states I generally find a way to kill off my first man and then play the second man. I also try to complete the level without actually finishing, like climbing the final ladder on elevators without taking the last step. That way I can also use the third and forth man with unique seeds for the randomize process.

[Shane_NC]

Ok thanks all for the tips I never really knew that.. One question though, I realize this can effect the rivets more profoundly than the pies, but do you think this effected the pies to any noticable degree? I mean since all fireball spawns in the same area and every PF besides some freebies seemed to be unique. Do you think the seed skewed my session? I went back and analyzed the tape after I heard the post and I has unsure really... any thoughts? Has anyone looked at the tape?

[homerwannabee]

You know it's cool to see variants of my 100 man analysis for Donkey Kong!


Not too sure if it was intentional or not, but still cool to see.  By the way, a 93 percent success on the Pie Factory is a really, really good passing rate.

[hchien]

Shane, I don't think it would have as big an effect on the pies, but from a purist standpoint I would say to save state from the end of the board before.  It also depends on how you play your pies.  Most people immediately run to the right and stop just short of the midline, in which case the fireballs will always behave the same during the time you are running.  For example, say you had a random seed which had the effect of no fireballs coming down the ladder during that time.  Then you'd be more likely to get a free pass from that save state.  Of course the moment you pause to wait for 5 fireballs to come out, it "randomizes" because you will always pause at a slightly different frame.

I've mentioned this before several times, but for those of you with a computer science background, it's because modern day computers are "finite state" machines.  That means there are only a limited number of internal states possible.  So given the same EXACT input, a finite state machine will always have the same output (ie it's deterministic).  Most modern computer systems use the current time as part of the random seed, so that has to be considered an "input" as well.  Briefly stated, there is no true randomness.

[marinomitch13]

Hank, wouldn't the fireballs randomize right away if you don't preemptively hold right? Could one just do that instead?

[hchien]

Well, it would be different, but not entering any input at all would also result in the same fireball behavior every time.  If you hit left/right/left/right really quickly and somewhat randomly, then you've essentially "randomized" it.

[Shane_NC]

Hey Hank, thats for the save state advice.. I tried your technique and managed to hit 44/50 pies trying your technique. This was actually my first 50 average last night from using the wrong technique. This works out to 88% which is not good , but I had 2 death that were pretty bad. 1 I got stuck on ladder due to 8 way joystick garbage and 2nd I rush and got hit my fireball at top with 3200 secs left. I still counted them and all in all a solid sesssion. Thanks for the tips I will round this into a 100 man analysis as well to hopefully bring up the 88%.. then I will average the 93 with the 2nd 100 man analysis to get an ever better idea. Im thinking easily over 90% effieciency.


http://www.twitch.tv/shane_nc/c/2880484

Thanks again
shane

[mikegmi2]

You guys just ruined the game for me.  Now I know that all the 1.1'ers have figured out the way to 'control the seeds', and complete barrel boards so that they won't ever get screwed on Rivets or PF screens.  And then they even let themselves get screwed at least once during a run to make sure nobody suspects or catches on to what they're doing.

On your bike,

</conspiracy>

The more you learn about the game, the less fun it becomes in a way.  They game has been really analyzed to the max at this point...to where we have hitbox charts, a full pixel perfect guide to the spring boards, code breakdowns of almost every aspect of the game (we know how freezers behave, rivet patterns, how the pies come out on PF stages, all the danger zones on barrel boards, probabilities of steering barrels, etc)...wild barrel hacks, MAME save states for super speed easy practicing...

Eliminating any human error, you're essentially 'going through the motions' in order to max out the game...and it's all up to the randomness of the game which will decide what your maxed out final score will be.  Now, or maybe it's already been figured out...the randomness isn't even random anymore, and could maybe be controlled?

Maybe after a while, someone will figure out a way to wait for the proper attract screen, hit start at the right time, and then follow a "perfect pattern" that will net you the max amount of points...based on following the proper perfect route for completing each screen...netting the max points...keeping the randomness essentially 'under control' on all barrel boards...and setting up the following PF or Rivet screen for no screwings...max smashes...etc.

[marinomitch13]

Mike, it's my understanding that the randomization happens at such a subtle level (hundredths of seconds in different timing here or there can affect it) that it's basically humanly impossible to attempt such a plot to get the 'easiest KS' or something like that.

[mikegmi2]

Yea you may be right but a long time ago nobody would believe you if you told them one day someone would score over 1,000,000 points on Donkey Kong.

They would say it's impossible.

[Shane_NC]

I went back and looked at all my deaths for both my KS games, 0 pie deaths in both games.

Kinda weird but each games deaths were on the same stages: 2 barrels , 1 rivet