NeuroShell Classifier Example

It is possible and many times profitable to use the NeuroShell Classifier to build classification models that can later execute in the NeuroShell Trader Professional or DayTrader Professional. If you want to read more about the benefits of purchasing NeuroShell Classifier to enhance your model building, read the section on called “Companion products”. In order to help you learn to interface with Classifier (NeuroShell Predictor is done roughly the same way), we offer this tip. Click here to download the NeuroShell Trader Pro chart, exported text file, and NeuroShell Classifier saved net. The net is called Classifier – you’ll need to put it into C:\NeuroShell Trader 5 for the chart to load properly. The chart has data saved in it already. The Classifier Example.csv file can go anywhere.

The best way to create Classifier training files is to insert inputs and an output into Trader Pro, then export a text file (.csv) from Trader Pro containing those inputs and output. You can, if you wish, first load the text file into Excel and perform any of the custom operations described in Companion Products.

If you load the NeuroShell Trader Pro chart you just downloaded into release 5.1 or better, you will see where we have created three inputs and an output called Class.

The inputs are just three traditional indicators – in this example we are not trying to explain how to make a great model, only how to perform Trader to Classifier to Trader mechanics. You can make other inputs in Excel later if you desire.

The output Class is 1 when the slope of prices over the next 3 days is positive (up) and -1 when the slope is down (negative). You can see the rule we used to generate Class. You don’t have to use slopes as the base of the output – you can use change in price tomorrow, or any other indicator that can be processed by a rule resulting in 2, 3 , 4 or so categories. You don’t even have to generate Class in Trader, you can make it in Excel later if that’s easier for you.

The three inputs and Class were exported to the text file Classifier Example.csv using the Tools->Export chart/data menu. If you load the file into Excel you will see the contents. Note the asterisks (*) in some columns and rows, which Trader inserts for missing data. In the current version of Classifier you will need to either delete rows which contain an asterisk in the Class column, or just exclude those rows from training in Classifier. Make any other changes you wish in Excel, like removing some bars if you feel they are outliers, selecting only certain ones, etc.

In Classifier you select the three input columns as inputs, of course, and select the Class column as the output. There several ways training can be accomplished in Classifier, but that is a different subject. We used the Genetic method, then saved the net.

Next we called the net from Trader, using an indicator in the category called “External Program and Library Calls”. We selected the probability of the category 1 (representing UP) as the probability to display. That is on the chart as “Classifier Signal”. (Note-If you do not own and have installed the NeuroShell RunTime Server, which is required to execute nets in NeuroShell Trader, you will not be able to see this Classifier Signal.”) We decided to call the probability true if it is > 0.5. You can examine the rule we used for that in the indicator we made called “Prediction of Class”.

Was this article helpful?

Related Articles