Results 1 - 10 of about 1,400,000,000 for self organizing map. (0.34 seconds)

Let me repeat: one BILLION four hundred millions... Oh.

A new version of Cortex is available. Same reg. codes.
New articles on the site: tutorials and ref. guide on language.

Please keep in mind, that some functions changed (minor), this means, you need to backup your data before updating.

Also, I have fixed (or so I think) some problems in FFBP scripts, results are now better.

Good luck.

P.S. SOM (Kohonen) are kind of a puzzle for mee, I am still not sure how to better use them. You can post suggestions here :)

Well... Cortex now supports plugins :)
A plugin is a third party DLL, and by following simple rules (see cortex.snowcron.com/scripting.htm#plugin) you can call this DLL's functions from Slang scripting language.

It allows you to create fast (written on low level languages) functions of your own.

If you have something to share (Cortex, not just anything :), just post it in this forum, with tags "cortex" and "plugin".

This is a new feature, and as usual, I appreciate comments and bug reports.

A known problem is the names conflict: if you have two plugins containing functions with same names, results are unpredictable. I am opened for suggestions.

In addition to things listed in an email below, I tested all scripts.

Hi all.
I have got some time, so I will revisit Cortex.

New version will be posted soon.

First bug I found was in the GENZ.NN in samples - the "reverse" checkbox was not set.
I think, the rest on NNs in samples may have similar problem

The idea of "reverse" is a bit obsolete, as it came from the "before scripting language" time. I was using Yahoo as the source of stock quotes, and they were providing data in the reverse order (first row of a table corresponding to latest data). The checkbox reverses the data array.

Note, that it only works for current tab in the NN dialog box, so if you want data to be reversed on the next tab, set the checkbox there as well.

Alternatively, use scripting to set your data straight, which means earlier data first and latest data later:

0
1
2
...

instead of

1000
999
...

----

Second bug fixed, that was preventing sample_nn_02.tsc from running.

----

Added validity check, so that we cannot use invalid file handle in file operations.

----

TABLE_LOADER used to look for start pattern equal to the whole line and end pattern CONTAINED in the line. Now both start and end patterns can be part of a line. Example: 12345 is part of 123456: match. Before, start line had to be 123456.

----

A properly working code for Timer is added: see script_ref.htm for details.

----

Bug fixed in forex_nn_01.tsc

----

Some bugs fixed. As usual, reg. code is the same as before.

Let's return back to one of the first posts in this blog. The one when I was suddenly - I had no idea! - told that someone took one of my Cortex samples, and created a commercial program based on it.

As you probably remember, I didn't mind, but I promised to watch their progress. I also mentioned, that it was just a simple code snippet, and it was never intended to make robust results.

Now, few month later, when the above mentioned trading system lost all it gained during the rally, I'd like to repeat my statement.

If you want to make a profit on a market, do your own research. There is nothing wrong with using someone else's developments, but you must - MUST! - understand how it works.

Yes, I am now sure that there is a silver bullet for the market. I have spent about 10 years experimenting, thought. And I am still not selling it :) because I want to make sure. And more sure.

Now, about the TS that these guys supposedly borrowed. It was using a predictive technique, created by a network with 15 neurons in the middle layer. What is the memory capacity of this network? What is it exactly, that it predicts?

Additionally, I did some research, that was not NN related, to simplify the task (it is all in a free article, on the site). But it means, that as soon as market changes, the system is GUARANTEED to break.

And so it did.

What a surprise!

A new, improved version is available for download. Reg. code is the same.

What is new?

Few bugs are fixed.
A new edition of the article for genetic computations with Cortex is available.

And the most important, a TRADE() function is added (see SLANG reference for details).

The way it works is the same that is already implemented in samples: it takes exchange rates and trading signals (plus sl/tp and lot sizes), and calculates profit. The only difference - it is a built-in SLANG function now, which means, you can do it FASTER. Plus, less code.

Before I sterted this forum, I had, what I am going to call "old Cortex". Then I created a new one.

The main difference (in addition to some bug fixes) is in the way weights are processed. They are different now, and accordingly, the MQL code that simulates the NN is different, too.

There are posts here, asking for old Cortex. Yes, you can use it (though I do not have it :) But please keep in mind, that a new one can do all the old one could. And more.

I strongly suggest, though not insist, that you study the scripts you use. If you are fluent with them, you may be able to find errors, and then the old Cortex will not be required.