Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Strategies & Market Trends : NeuroStock -- Ignore unavailable to you. Want to Upgrade?

To: Len Giammetta who wrote (65)	9/17/1998 10:21:00 PM
From: Jay Hartzok	Respond to of 805

That's exactly what I'm saying, but you can't enter just the target stock symbol. The program won't let you because all entries are inputs and Neuro will demand some kind of data for the entry. If you try it, you will get an error message. Jay

To: Len Giammetta who wrote (65)	9/17/1998 10:58:00 PM
From: Jay Hartzok	Read Replies (1) \| Respond to of 805

Len, All this talk of verification is getting me confused and I don't like being confused. I still say that verification is already built into the program, but to satisfy my doubts I feel that I have to test the net on data that the software does not have available, just to make certain that it can't cheat. Read the following excerpt from the FAQ thoroughly, and see if you can see why I think verification is already built in and done automatically. _____________________________________________________ There are two phases or steps in every neural network program: training and application, this has been documented in countless books and papers in the subject. The first step is the training phase, here the network formulates a model or hypothesis of the system under analysis (in this case the future price of a stock based in present and past data), then the model is tested using a lot of historical data. For each day in the available history the neural network makes a prediction based strictly on information prior to that day, then the prediction is compared to what actually happen in the following ten days or so of the date under test; if the prediction was correct then the model is reinforced, if it wasn't then the model is corrected. Each round of tests is called an "epoch". After enough epochs, the resulting model is a very refined representation of the stock's behavior. No model is ever perfect because there are just too many factors influencing a stock's price, but we know that the complexity of the model is related to the number of neurons in the network; a network with more neurons is able to refine its model more and give more accurate predictions. The second step is the application, in this case the prediction of a stock's future value, and a buy/hold/sell recommendation. Here the neural network uses ALL required information, down to the last day, and feeds it to its model of the stock under analysis. The resulting recommendation is based in what the network predicts the stock will do in the following ten days from the last date in the historical file. The network predicts the future performance of the stock based by feeding the latest data to the model it refined during training. So you see, during training the network needs to know the result of the prediction so it can refine its model, that is why you can't train to the latest day since its result in not known yet. When you hit the PREDICT button the network automatically uses ALL the data it needs to predict the stock's behavior including the latest data, not the dates in the training dialog box.