Can Braincel find the best
set of inputs ?
Yes. Braincel has a BESTNET
option that directs the program to repeatedly eliminate the least useful inputs
until all is left are the most useful. Braincel knows that the only way to do
this correctly is to have two data sets, one for training the net and one for
testing.
You can also specify a minimum number of
inputs that the neural net must have.

Can Braincel find the best
neural net size and shape ?
Yes. Braincel has a BESTNET
option that the user can direct to find the best number of layers and number of
processing modules (neurons) at each layer.

What's the best application
of a neural net in a trading system ?
Numerous users of neural net
technology have reported that using a neural net to predict the future TRADING
RANGE of a price time series is likely to produce better results than most
other kinds of forecasts, including future percentpricechange, future price,
and future volatility.
To forecast future price range, you need to
create TWO neural nets. One to forecast the HIGHEST price of the next N days
(minutes, ticks) and one to forecast the LOWEST price of the next N days.

Does Braincel automatically
normalize input data?
Yes. Braincel will normalize each
input column separately. In general, normalization is a 2step process that
shifts and scales the data in each column into zscores. That is, normalization
evaluates the mean and standard deviation of all values in a column, then
applies the following formula to each value X in the column:
Z=(X  mean) / Stdev
The normalization process in Braincel is a
little more sophisticated than the above formula, producing similar results
that are less influenced by outliers.

Can Braincel operate on data
in realtime?
Yes. Braincel can process data in
realtime in Microsoft Excel and TradeStation.

How often do I have to stop
to retrain my neural net?
A properly trained and tested
neural net does not need to be retrained every day. You may want to consider
retraining on a weekly or monthly basis. It all depends on how fast long term
market behavior changes in a month.

What's the best way to
increase a NN's accuracy?
The training session
automatically sequences through data records in random order. As a consequence,
different training sessions will likely produce NN models with different
weightings. You can get stable and accurate results by creating 5 NN models
(using the same data and training parameters for each model) and using the
"Olympic Scoring" method: after running a data record through all 5
models, eliminate the highest and lowest result, then average the remaining
three values.

What programming language is
needed?
VBA in Excel or Easy Language in
TradeStation. For each platform, sample code is provided to help get you
started.

Which platforms allow
creating NN models?
You can create NN models in
Excel. TradeStation users would need to first create a Braincel NN in Excel,
then use Easy Language to apply the NN to realtime or historical data.

