| |
 |
|
|
Science Forum Index » Compression Forum » indicators of a good compression ratio
Page 1 of 1
|
| Author |
Message |
| novice |
 Posted: Tue Apr 22, 2008 3:26 am |
|
|
|
Guest
|
Hello I have a algorithm in the works but before I fully implement it
and find out its not good (atleast nothing top of the line) what are
good metrics in telling me that I have a good algorithm.
i.e. My idea is a dictionary approuch (lossless) and if I can tell
each address used is compressing 3-5 bits a "use" is that good? For a
simple huffman does it compress 1-3 bits per address use? LZW? others?
I ask because 3-5 bits seem small, but thats per address use, which
means that could be 3-5 million addresses used in even small files -
which means 9 - 25 million bits compressed.
Perhaps its not a set number like 3-5 but perhaps a percentage?
I am aware that some address wont compress but expand -
Thanks. |
|
|
| Back to top |
|
| Phil Carmody |
| Posted: Tue Apr 22, 2008 11:47 am |
|
|
|
Guest
|
novice <Tjackson.1982@gmail.com> writes:
Quote: Hello I have a algorithm in the works but before I fully implement it
and find out its not good (atleast nothing top of the line) what are
good metrics in telling me that I have a good algorithm.
One thing that's absolutely _essential_ is that your algorithm
must expand the majority of its possible inputs. If you can't
achieve that, you're on a non-starter.
Quote: i.e. My idea is a dictionary approuch (lossless) and if I can tell
each address used is compressing 3-5 bits a "use" is that good? For a
simple huffman does it compress 1-3 bits per address use? LZW? others?
I ask because 3-5 bits seem small, but thats per address use, which
means that could be 3-5 million addresses used in even small files -
which means 9 - 25 million bits compressed.
Perhaps its not a set number like 3-5 but perhaps a percentage?
I am aware that some address wont compress but expand -
It all depends on the input data. Without information
on the input data, it's impossible to say anything specific.
(And I'm not sure what you mean by "address" here.)
Phil
--
Dear aunt, let's set so double the killer delete select all.
-- Microsoft voice recognition live demonstration |
|
|
| Back to top |
|
| Thomas Richter |
| Posted: Wed Apr 23, 2008 2:16 am |
|
|
|
Guest
|
novice schrieb:
Quote: Hello I have a algorithm in the works but before I fully implement it
and find out its not good (atleast nothing top of the line) what are
good metrics in telling me that I have a good algorithm.
Compression ratio on data you consider typical for your application?
Quote: i.e. My idea is a dictionary approuch (lossless) and if I can tell
each address used is compressing 3-5 bits a "use" is that good?
Hard to say without really having data on *what* you want to compress,
i.e. what is its probability distrotion? What is an "address"?
What is the entropy (under your model) of the source? How close do you
get to it?
Quote: For a
simple huffman does it compress 1-3 bits per address use? LZW? others?
I ask because 3-5 bits seem small, but thats per address use, which
means that could be 3-5 million addresses used in even small files -
which means 9 - 25 million bits compressed.
Perhaps its not a set number like 3-5 but perhaps a percentage?
Perhaps it's unclear because you haven't said what you're doing? (-;
So long,
Thomas |
|
|
| Back to top |
|
| |
|
Page 1 of 1
All times are GMT - 5 Hours
The time now is Fri Sep 05, 2008 12:53 am
|
|