Compression - Need some help interpreting results from new compression method

Author

Need some help interpreting results from new compression method

Einstein

2007-11-22, 7:56 am

Normally I don't like to come here, a couple people here are not nice
people, but regardless I am trying to figure a way to create a
statistical occurrence table and check for compressibility based upon
the results I have.

I have 2 different possible outcomes with an equal occurrence chance
for each outcome statistically. This much I have proven, but stress,
fatigue, and 3 jobs is killing my ability to try to proceed a little
further.... thank God one of the jobs ends next w

....

Anyhow each outcome might seem to result in gibberish, but my method
actually has them sorted successfully. Guessing wont be useful, I just
wont reply unless you somehow get an exact dead on of my method. Lets
just say I want to keep it in the closet just a little more longer.
But no loss is involved, some minor growth is included, but
statistically it is extremely low.

RESULT #1
00
01
0
0
1
1
1
1

Result #2
000
001
00
01
01
01
0
1
1
1
1
1
1
1

While they are balanced in total 1's versus 0's in each method, it
seems to me that the longer length of some of the outcomes means we
might be able to obtain compression. It "feels" as if there is
patterns attackable with one of a variety of methods, yet proving it
is a bit beyond my time/energy at this moment.

Anyone care to assist me with this? btw result #3, with this system
was funny, 0 and 1.. exactly :P

I have considered re-rerunning my system, have considered using
variants, but I would love to see the results from this if anyone has
spare time and some curiosity.

Providence

2007-11-23, 6:56 pm

On Nov 22, 5:38 am, Einstein <michae...@gmail.com> wrote:
> Normally I don't like to come here, a couple people here are not nice
> people, but regardless I am trying to figure a way to create a
> statistical occurrence table and check for compressibility based upon
> the results I have.
>
> I have 2 different possible outcomes with an equal occurrence chance
> for each outcome statistically. This much I have proven, but stress,
> fatigue, and 3 jobs is killing my ability to try to proceed a little
> further.... thank God one of the jobs ends next w

....
>
> Anyhow each outcome might seem to result in gibberish, but my method
> actually has them sorted successfully. Guessing wont be useful, I just
> wont reply unless you somehow get an exact dead on of my method. Lets
> just say I want to keep it in the closet just a little more longer.
> But no loss is involved, some minor growth is included, but
> statistically it is extremely low.
>
> RESULT #1
> 00
> 01
> 0
> 0
> 1
> 1
> 1
> 1
>
> Result #2
> 000
> 001
> 00
> 01
> 01
> 01
> 0
> 1
> 1
> 1
> 1
> 1
> 1
> 1
>
> While they are balanced in total 1's versus 0's in each method, it
> seems to me that the longer length of some of the outcomes means we
> might be able to obtain compression. It "feels" as if there is
> patterns attackable with one of a variety of methods, yet proving it
> is a bit beyond my time/energy at this moment.
>
> Anyone care to assist me with this? btw result #3, with this system
> was funny, 0 and 1.. exactly :P
>
> I have considered re-rerunning my system, have considered using
> variants, but I would love to see the results from this if anyone has
> spare time and some curiosity.

hi there, I've been analyzing the method you're presenting, and as
you're doing statistical analysis per bit, you'll end up with data
corruption, with the probability being, of course, the inverse of the
successful prediction ratio.

if you boost the prediction ratio up, awesome, but consider that 99%
success means 1 of 100 bits will be corrupted. if the corrupted bit
cannot be specifically marked as corrupted (increasing the total
transmit bit count by 100%) 100% of the time, then all the sudden
you're no longer doing lossless encoding, and are instead working with
lossy formula.

without a 100% method of lossless encoding, you aren't going to be
able to fold the data layers over top of each other, thus you lose the
composite construction, blasting your n'th dimensional method back to
2d, and a lossy 2d at that.

by no means am I proposing that your research is fruitless, as you'll
come to really interesting conclusions and starting points for other,
more intricate and sublime methodologies following from understanding
your path here. however, it's your projections into the static data
stream that allows you to fill in the missing information, and this
isn't taking place at a quantum level. figure out how to properly
*generate* the curves that you need to align the statistical mechanics
with the patterns you're seeing, and you'll be one step closer to your
goal.

chris.

Einstein

2007-11-23, 6:56 pm

Thanks for replying Chris, and for the thought out post.

However fortunately this is not a predictive stream, this is actual
result outcomes from the system I have set up. I will say this is only
a portion of the input, the rest is sitting in an identified length
subsection and is not being compressed 'at this moment'. This factual
outcome results is the unique aspect of my system. I need an actual
probabilities table written up on this asap. Sadly no time :(

Providence

2007-11-24, 6:56 pm

On Nov 23, 2:44 pm, Einstein <michae...@gmail.com> wrote:
> Thanks for replying Chris, and for the thought out post.
>
> However fortunately this is not a predictive stream, this is actual
> result outcomes from the system I have set up. I will say this is only
> a portion of the input, the rest is sitting in an identified length
> subsection and is not being compressed 'at this moment'. This factual
> outcome results is the unique aspect of my system. I need an actual
> probabilities table written up on this asap. Sadly no time :(

the next gate is the injection portion.. it would seem to be possible
to properly justify the positional portions of compressed and
uncompressed sections, yet you're looking at it lkie tihs. as yuor
mnid can poreprly deodce the positions of the data from a permuted
source, intuitively it'll seem like this is also possible at the data
compression end.

the actuality is not so much, unless again you're able to isolate the
curves that are providing the rearrangement.

all this combined means that if you have to specifically mark the
portions as being 'compressed' and 'uncompressed', then you'll have to
use block sizes of much greater than a single bit/byte. at which point
you need to make sure that the
data is able to be aligned without losing all the benefits from the
compression method.

just be aware that your mind is capable of 'livening up' any data
source by providing fractal cohesion, and this is the human element.
as soon as you reduce the human element out of the equation by using a
fully permuted data stream of k elements in base t, you have to really
understand why some things seem intuitively possible, probably,
likely, and necessary.

i think the reason you're getting improper reflections from this group
is that your system isn't fractally cohesive.. what you have going on
simply doesn't work losslessly, and this is a lossless compression
forum.

post the logical arguments, say exactly the position you're at making
this thing work, and we can probably figure out if it's possible, and
learn a lot in the process even if it's not. if you're totally honest
with your approach, you'll find it a lot easier to get a true look at
where your system is loosing cohesion.

chris.

Einstein

2007-11-24, 6:56 pm

Sir I can, and would, put a $10,000 bet that this is the outcome of
the method "Statistically". The math on that is irrefutable.
Additionally I can, and would place a bet of $10,000 that it is
lossless.

I will not be posting methods til I have had time to check the results
out, determine the statistical chances of compression, and and see if
the level of compression on this portion outweighs the increase in
size needed to separate the portion out. Sadly I just had 3 more
projects dropped on my lap with a required to be finished by Monday...
on top of a pre-existing 3 projects.

And before you ask sir, yes it is possible to return the data entirely
to the original code with no loss.

I will write out a basic version of the statistical layout when I can
later. Every outcome of 2 groups. I am desiring every of say 16
possible combinations in sequence, which would of course be the best
statistical model to generate a probabilities table off of, and
therefore to also start examining deviations from the statistical
norm. This of course would be a rather large outcome, but the
information gathered would be priceless for understanding if there
could be compression based upon the results therein.

Einstein

2007-11-25, 7:56 am

ok time to do some basics

0000
0001
000
000
001
001
001
001
0100
0101
010
010
011
011
011
011
000
001
00
00
01
01
01
01
000
001
00
00
01
01
01
01
100
101
10
10
11
11
11
11
100
101
10
10
11
11
11
11
100
101
10
10
11
11
11
11
100
101
10
10
11
11
11
11

there are a simple version of 'statistical outcomes' if we placed a
random pair next to each other, using the 8 outcome version.

0000 x 1
0001 x 1
0100 x 1
0101 x 1
010 x 2
000 x 4
011 x 4
100 x 4
101 x 4
00 x 4
001 x 6
01 x 8
10 x 8
11 x 16

64 total outcomes when viewed as direct results, however when we look
at 00, 01, 10, 11 results... well we can't... statistically we don't
have a large enough sample if we wish to just run it as existing
length, unless we add a system to account for our non-even length
issues. However every time I try to make a model to account for that,
I have issues. So my goal is to run it at greater than 16 bits in
length and to just do a lesser version to account for the non-even
lengths. I am probably missing something rather easy to do to generate
the needed information with a basic series, but I am so tired anymore,
so much work... this is the sort of help I was hoping to find, so far
5 usenets and no luck. Of course each result is a probability, and I
am at this moment looking for statistical outcomes.

Mostly I am relying upon the facts that in very large files the total
number of 1's and 0's tends to be a neutral amount, within a specific
identifiable range, thus I can count that 'usually' I should have a
fairly even disposition of any possible string size up to 10 bits in
length among all possibilities of that string, within a certain
allowable margin of error.

Once knowing the statistical 'fallout' after engaging my algorithm I
would know each individual lines compressibility, chance of occurence,
and then be able to draw a conclusion about if it is compressible,
under what methods it is compressible, and if it is worth trying due
to the minor size increase I experience to track initial file size and
the size after the algorithm has enacted, and 2 bits to identify no
algorithm, version 1, version 2, or version 3 (3 being the one which
comes out statistically tied). Also the results will tell me if I need
to work with a lower, or higher base inside the algorithm for a
specific part of the formula for the algorithm.

Providence

2007-11-27, 9:56 pm

> I will not be posting methods til I have had time to check the results
> out, determine the statistical chances of compression, and and see if
> the level of compression on this portion outweighs the increase in
> size needed to separate the portion out.

indeed, this is what i'm talking about.

in experiments of my own carried out to their full, this was the final
integration that, at best, led me to equality, zero sum.

i would suggest thinking very deeply about why exactly you're needing
to seperate portions out, instead of creating a pointer to the data
from a generator curve.. you're capable of this level of insight if
you've got this far.

consider how quantization works. check out the bases operations on
each other. it's quite neat, and they're capable of providing plenty
of data with very little set up, you just need to allocate a
logarithmic pointer into the stream.

:)

chris.

Sponsored Links