|
ANNI's Frequently Asked QuestionsLast modified: Sunday, June 24, 2007 How do I use ANNI's Outputs to help with my investment strategies? Are there other costs involved to use ANNI? Why does ANNI require 675 entries to train on? Can ANNI handle training with 30-50 stocks and mutual funds? Can you export ANNI's forecasts? Does ANNI support MetaStock Data? How do I use a custom data file in ANNI? Does ANNI automatically update the data files? Why does the close values appear to be off on some stock? Should I retrain ANNI after updating the data file? Can ANNI be opened twice at the same time to run two files simultaneously? Why does the Calculate button do nothing in the "Chance of Success" dialog? Does ANNI train faster when minimized? Why are my high, low, open and volume indicators sometimes not available? Why is the Fundamentals tab not always shown? What does "Linked Training" mean? Why does ANNI keep telling me it can't find xxxx.nni file when it starts? Why does ANNI's keep training forever? Why are there not as many data entries available in ANNI as there are in the file? Report companies that do not respond to questions when emailing their Investor Relations department Are there other costs involved to use ANNI? Once ANNI is purchased there are no other costs involved to run the majority of components in ANNI. This includes integrated retrieval of free end of day historical data for all securities Yahoo Finance supports as well as other information ANNI retrieves from free websites. There are additional services you can subscribe to however that can enhance ANNI considerably depending on the type of trader you are and how you plan to use ANNI. Minor upgrades are also free for life along with low costs upgrades to major versions. I've trained for many generations and ANNI still shows a low confidence level. Is there anything I can do to improve this? ANNI is an extremely dynamic and robust securities modeler which can overcome many complex problems. Sometimes though ANNI may not perform as well as expected. There are several reasons this can occur
The bottom line is that ANNI does not always work on every security (something that should always be kept in mind). ANNI provides tools to help PreScreen a stock to determine if it will be suitable in ANNI in addition to the Summation View providing flags in case any of the above conditions are violated. But keep in mind that if after trying the above solutions do not work, do not try and force it to work or use sub-optimal results in the hopes that it will suddenly begin to work. You are much better off finding another security that will perform better with ANNI. Why does ANNI require 675 entries to train on? When working with Neural Nets it is important that you have sufficient
entries for the Neural Nets to learn off of. The data that is learned is
called the solution domain or in simple words, where the Neural Nets will
perform best in. If a system is forced to predict outside of this range,
extrapolation occurs and the accuracy of the result are severely restricted
(since it's predicting in an area it doesn't have experience with). This
problem is compounded when insufficient data is provided to the system since the
system now becomes highly biased towards a narrow solution domain. Since
ANNI also uses various indicators for inputs, up to 200 additional entries are
required to buffer the calculation of these indicators (for a total of 675
entries). Therefore in order to protect a user from accidentally creating
a system that is highly biased that can also be forced to extrapolate its
predictions, a minimum amount of entries is required to train ANNI's prediction
system and Basic Trade System. This is obviously a better solution then
ignoring this important aspect of Neural Nets and allowing someone to use
results that could be no more accurate then flipping a coin. Can ANNI handle training with 30-50 stocks and mutual funds? ANNI is specifically designed to work with large Portfolios (as high as 125 stocks - it can go higher but some limitations can occur at this point especially with slow computers). Because of the limitations of training though, ANNI files must be trained one at a time. If all 125 stocks need to be tracked simultaneously you will want to look into the Real-Time Prediction System which can be ran on all stocks simultaneously with minimal impact on CPU resources. This is great to use during the market and while the market is closed used the more in depth ANNI files to give you an edge at the beginning of tomorrows trading. Can you export ANNI's forecasts? ANNI's exporting features are quite extensive with just about everything you see having the capability to be exported in some way or another. You can export the Prediction System's outputs, the Trade System's outputs, the Fundamental Information, Analysis Information, Portfolio Reports, Portfolio Snapshots...even Graphs can be exported! Does ANNI support MetaStock Data? v2.16 and above have full direct MetaStock Data support. This includes the newer XMASTER format using .MWD files. For details on how to use MetaStock data with ANNI simply click here. Please note that ANNI will not update MetaStock data files though, it will only read them. How do I use a custom data file in ANNI? If you do this often, follow the second method. Otherwise this is the easiest method. Quick and dirty, easy method 1:
A little more involved, but if you do this often this is the way to go:
For more details on creating a custom file for use in ANNI, visit Advanced Tips and Tricks Does ANNI automatically update the data files? ANNI automatically downloads the data file during initial creation. After creation the data file is automatically kept up-to-date as long as this option is selected (the default is on). This option can be changed by going into the Data Retrieval configurations and changing the "Automatically Keep Data Up-To-Date" to suit your needs. In cases were someone is doing in depth back testing they may want to turn this feature off. Why does the close values appear to be off on some stocks? When stocks have a split or cash dividend, it can directly effect the stocks
price. If you where to view a chart that did not use adjusted close
values, you usually would see a sharp change in price at that time. As
humans, this is easily understood. For neural nets, this can cause some
problems. Therefore whenever a stock's data file is read that contains an
Adjusted Close column, this value is used rather than the actual close value
itself. By doing this, the neural nets can pick up on the inherent cycles
much easier than they would have without it. What this means for the user
is that any displayed close values will be the value of the adjusted close
rather than the actual close at that time. If you open up the data file
directly, you can see this "Adj* Close" column and it's values for
verification. Should I retrain ANNI after updating the data file? The system should continue training for at least one complete generation prior to evaluating the new results otherwise the systems will eventually diverge especially if a new trend is taking place. This can usually be seen by a sudden decrease in the performance (confidence level) after new data was downloaded. This is because the current system has begun to diverge and no longer represents the current market conditions. If this divergence is significant enough, it may be advantageous to reset the system completely...otherwise continue training as normal since the new data will only enhance the existing learned trend. Can ANNI be opened twice at the same time to run two files simultaneously? ANNI is designed to efficiently run with
only one instance and will utilize a Multi-CPU, Multi-Core or HyperThreaded CPU
computer completely if both the Prediction System and the Trade System are
training. Running more then one instance of ANNI will cause your CPU
resources to be split up possibly not providing any advantages over running just
one instance. But if you need to, you can without any problems (assuming
you are not opening and training on the same files otherwise you may eventually
get a file access error of course). Why does the Calculate button do nothing in the "Chance of Success" dialog? If you use the sliders, the values change and get recalculated on the fly. If you manually type in a value in the edit boxes...that's when you would select the Calculate button to get the results updated. The sliders are by far the easiest to use. Does ANNI train faster when minimized? Yes, when ANNI is minimized, no graphics, displays or text are updated. With v3.xx though a complete change on how the views are managed has taken place reducing the impact any single or group of views will have on training. Still, you can get a few more CPU cycles for training if you do keep it minimized. Why are my high, low, open and volume indicators sometimes not available? If a file is recognized as a mutual fund (and some indexes) all high, low, open and volume indicators will not be shown (since they are not available...or irrelevant). This can also occur if there are errors in the data file such that it appears like a mutual fund or index (in that case you need to either redownload the file or locate and correct the errors in the data file). Why is the Fundamentals View not always shown? Mutual Funds, Indexes and Foreign securities do not have Fundamental information. Therefore this view is hidden when these file types are open preventing any confusion on the information supplied in this view. What does "Linked Training" mean? Linked Training is merely for starting and stopping training of both the Prediction System and the Trade System simultaneously. In no way does the training process ever intermingle just because this is selected. This feature is just for convenience. Why does ANNI keep telling me it can't find xxxx.nni file when it starts? ANNI's default settings are to open the last file it was running with when it starts up. If you delete or move this file, every time ANNI starts it will say it cannot find it. Simply create a new ANNI file, or after ANNI is open go into the Configurations/Startup Options dialog and unselect the "Reload Last File on Program Startup". Why does ANNI's training keep training forever? In order to train groups of files more efficiently, ANNI implements some early stopping features. This way during batch training, the files that need the most training time will get it. By default one of these early stops is selected (the confidence level threshold). To change this default behavior, go into Configurations/Advanced Training Options and change the options in the "Stop System Training When..." group located at the bottom. These settings are only checked at the end of a generation pass and not during (this way you can go through at least one complete generation before it is activated). Why are there not as many data entries available in ANNI as there are in the file? ANNI uses technical indicators and time series inputs. ANNI needs to have enough days cushion to allow for the max days required for any technical indicator - 212+ days (at least the limits ANNI uses). In addition, ANNI can use up to 75 days input for one net - not 1 day (and in some cases 150 - not public though). So it takes 212 + 150 = 362+ max trading days additional needed. This is subtracted from the total entries available to determine how far you can slide the date slider over. The buffered trading days could be less for any given condition, but it is not dynamically calculated because ANNI uses so many optimized settings that it can't be determined on the fly (else data would have to be reloaded every time it changed). But it can be guaranteed if it is kept constant at the max. So when you slide the date slider over to the end, technically you are using those additional dates to some degree. But the results produced are only going to be for the date the slider shows since ANNI can take all of the prior dates as inputs just to predict tomorrows value. If it went to far back, ANNI would not have any inputs available in order to compute an output (technically the program would blow up when it accessed memory that didn't exist). |
|
Home | Download | Register | Buy | Current Issues | Contact Us | Join Associates | Help |
Copyright © 2008 Neural Science