ANNI's Frequently Asked Questions

If the best nets disagree on direction when viewing them in the large graph, does this mean it is not a good system?

The first and foremost reason is usually not enough training. If you haven't completed at least 15 generations (or a few new generations after the data file has been updated), continue to train before further analysis.

On the contrary, it means the system is doing what it's suppose to. What ANNI is basically doing is picking up on multiple signals rather than one, which can obviously vary. When using ANNI's output though, only the weighted average output should be used. The Best Nets option is to allow detailed analysis on the success for each net in the Test, Train and Eval periods. Trying to compare them in the prediction period could be rather hard since each one represents something slightly different. Continue reading for more detailed information...

This is one of the most powerful features in ANNI. A neural net when training will train towards the strongest signal first. If someone were to use 1 neural net, then they would see the biased output of the strongest signal (with smaller and smaller ones underlying it). The problem of course with stocks is that they have many signals meshed together to form the final output…the closing price. Understanding the simple concept of moving averages we find that we can look at different cycles while reducing the noise of the other signals. Because ANNI can look at several days simultaneously (have varying input periods in other words), we are giving that same power to ANNI. So we can look at the strongest signal for a 3 day period and the strongest signal for a 7 day period and...etc. Then we can combine them, just like the stock market does, to produce a final price (the “Weighted Avg” method). So when you view the Best Nets, your looking at all the different “best” signals for their given input period which can vary a little or sometimes a lot depending on the security being analyzed.

When viewing all the best nets in the large graph, look at the intensity of the green lines. The brightest green line is the one ANNI feels the most confident about (it's the top net basically). The darkest green line is at the bottom of the best nets. This is important in understanding ANNI's "Weighted Average Output". Weighted meaning the top nets have more say so than the worst nets. So for nets that disagree a lot and have low confidence...their say so into the final weighted output will be very small. Now you can see why the weighted average output works best and should always be used as ANNI's final output - it enforces the best, cushions ones that could be potentially wrong and in doing so creates a more robust system

For example, if only short term traders where driving a particular stock currently the stock would be going up (so one of ANNI's nets that honed in on this signal shows this). But if only long term traders where driving this same stock currently the stock would be going down (so one of ANNI's nets shows this as well...making it appear they have mixed opinions - and technically they do). Combining both traders simultaneously (which is what the market is really currently doing) causes the stock to remain in some reciprocating middle ground (a stocks true value).

To understand these methodologies in more depth, you will need to research digital signal processing, Fourier Transformations, Wavelets and Nyquist Frequencies. This will explain how multiple signals can be extracted and analyzed separately...and later recombined (and where each signal could differ greatly between themselves). Although ANNI does not specifically use Fourier or Wavelet transforms for it's inputs (it may soon though), the principals and techniques are similar in concept.

Back