GoldenGem® and StockDownloader
Return to Home Page
User instructions
Technical specifications
Important announcement
Download links
Optional configuring
Is it free?
For verification that it works and other more detailed data go to the UK site
http://www.goldengem.co.uk
Goldengem is a state of the art general purpose Neural Network. It is a simple no-nonsense program, containing
a very sophisticated mathematical calculation, which has been very successfully used by a few top traders.
Availability:
We have obtained a version which is free until July 1, 2006 Download Link
It can also be found on 
Stock Downloader
GoldenGem comes bundled with StockDownloader, an automatic utility which will download and update (for free) the current
two years of price and volume data of any selection of tickers. To use StockDownloader, enter tickers into the text
window and press the Download button. See StockDownloader Information
for further information. It is important to remember that GoldenGem looks at not only the graph you are displaying, but all loaded
prices and volumes. It would be a mistake to use StockDownloader to download just a single ticker, you need to include
the tickers you think may affect the one you want to look at.
User instructions
If you use StockDownloader, the volume and price information will be loaded automatically.
If you wish to just use a data file instead you will find that
the program can read ascii data files such as the sp500hst.txt file, which is the last 12 months of stock prices and volumes
from the Standard and Poor index, it is included and can be found free of charge on http://biz.swcp.com/stocks/ .
Goldengem is completely flexible regarding file formats
Click 'load' and 'run' it will start running immediately. The load window in the screenshot contains the top ten Standard and Poor stocks, for the last 12 months, and you may type others before you press load. For example you could add to the list one which you think may be affected by prices and volumes of the top ten. Then
you can press run
and watch the graphs progressing in historical time. A neural net is not a replacement for human intelligence.
You must know in advance which tickers really are related to the objective price, either because they affect it
or act as a 'weathervane' by reacting more quickly to trends that will later affect the objective price. The displayed choice of the top ten
could lead to incorrect or misleading answers if you do not know how to replace it with a choice of your own.
Each blue trace is an advance copy of the corresponding red trace, advanced by a number of days on the slider. The green curve looks at all loaded volumes and prices, at the time indicated by the earlier red trace, and learns to match the advanced blue trace. At the beginning, you will want to have the sensitity high, as it will want to be learning fast. As it runs, you see the traces going across. If the settings are correct,
then each in each iteration the green curve will match the blue curve better, because it learns more about the best way to combine these to try to copy the blue curve based on all the data it sees.
After a few minutes you should set the sensitivity a little lower and let it run for several minutes. You may see the green curve has learned to match the
advanced blue curve. Then turn sensitivity to zero. If the green curve continues to match the advanced blue curve, then you are seeing a learned relationship: the green curve is no longer seeing any information about the advanced blue curve. In the rare cases when this happens, this means, the green curve may have learned how to predict the advance copy of the red curve by looking at all past effects. It can continue the calculation up until the current time, and give an opinion about an equal number of days yet to come. You see this in the first screenshot as the part of the green curve that extends past the verticical blue line. The screenshot is the ninth iteration, where the prediction would not be reliable. [note added May 5: the screenshot really was uploaded on April 19, and the predicted graph does match the actual graph of XOM but this was a coincidence as the choice of top ten S&P is not likely to include stocks that respond more quickly to market trends than XOM, nor certainly is a large oil stock going to be influenced by these.]
Optional Setting Up and Configuring
The default settings are usually the recommended settings. It is possible to choose the number of stages, and neurons in each stage. The number of neurons in the first
stage is twice the number of tickers you have selected (because both volume and close are looked at by the
first stage), and the number of neurons in the last stage is just the number of tickers (because it is trying to
match the close price).
These numbers are set automatically when the tickers are loaded.
The default configuration has three stages, with 64 neurons in the intermediate stage so it is not necessary
to use the configuration button. The number of stages
can be increased to four, and the number of neurons in the middle stages can be increased up to 256; and the net will be more intelligent but on most computers very slow, and it will take a lot of work to get it to
settle down, runnng it and adjusting the sensitivity, and a lot of waiting.
Registration:
The program is written by a Math professor in England. If you decide to keep it
you will eventually need to register the program. The permanent registration key comes directly from England. It comes with essentially
unlimited support, and the license to obtain all future upgrades.
When you register, we request a nominal charge
(currently 75 US Dollars), to cover future development costs. We feel this is fair as holders of a permanent key will certainly benefit
from both past and future development. If for any reason you
aren't able to make this payment, contact the email address on your console or shown on the console in the screenshot and
you will receive your permanent registration key anyway. If you are a student using the program for study, you may
receive a free registration key if you write from your University email address.
As soon as you have given your email address to Paypal, check your email. Your registration key will be there.
UK business directory
Here is another neural network site (snowseed)
uk business directory
Announcement:
A new feature has been added on 17 May 2006 (an additional bias neuron in level 1) which
improves the performance tremendously. You are welcome to update your copy of GoldenGem from this site if
you already have one. You can determine whether you have the old or new version by looking at the number
of neurons in level 1 when a data file is loaded. The former version of GoldenGem has 2n where n is the
number of tickers loaded. The new version has one more (2n+1) so is a very slightly larger configuration,
but with a vast improvement. Your registration key will work to permanently register the new version. Thank you.
Data Recovery Software