Re: [computer-go] Data Mining

["Frank de Groot" <frank@xxxxxxxxxxxxx>]

> So awaiting the future for you to get 256 terabyte,
> we must assume your 'learning' works well?

How it works, is that I make a matrix of 8 million x 8 million patterns.

I record for each pattern, how often it overrides another pattern in a
position, and how often it doesn't, for each pattern.

That's 256 TB when I use 16 bits for a value.
When it's all done, I calculate the least error values for each patterns.

But, as I said, I found a way to do it in 400 MB RAM.


> Your last name is rather dutch.

Perhaps that's because I am Dutch.


> There is a government nation wide organisation called NWO.

The New World Order?


> And no you don't need hundreds of terabyte in order to create 8 million
> patterns.


No, but to cross-correlate all of them against eachother the fastest,
ideally needs a matrix of them in RAM.

2.8 x 10^14 bytes.

But there are ways to do it slower with the same result.

_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/