Built-in Rule Sets

Back Home Up


Yahoo Charts


US Markets Daily


Forthcoming Releases:

Fuzzy Logic:

Neural Net:
Release of NXL3


Built-in Rule Sets

"3x3 Shape" & "5x5 Shape"

As a reminder, the 3x3 Shape means that the 2 variables will both use a 3-state fuzzy set (such as "Low", "Medium", and "High").  In the same way, the 5x5 Shape means each variable uses a 5-state fuzzy set.

The sFLC3 API includes a number of predefined rule set models to facilitate the building of fuzzy logic controllers.  This should provide enough configurations to cover most problem solving cases in trading, and while one of the primary objective was simplicity, versatility comes second so rule sets can be fully customised.

Secondly, one must keep in mind that our rule sets are "3D cubes".  The 2 inputs provide the base surface (3x3 or 5x5) and the output is the vertical dimension.  We want to reassure you straightaway: our rule sets are not quite "fancy Rubik's cubes" yet, but we can indeed look at them in different ways.  For instance, if we forget for a minute the semantics behind variables, nothing stops us from inverting each variable (high becomes low and vice versa) hence the same base model can be derived in 8 different rule sets (2 times 2 times 2).

The purpose is certainly not to complicate our models unnecessarily, but on the contrary to provide more versatility to be used in an automatic FL controller optimisation using Genetic Algorithms for instance.

3x3 Predefined Rule Sets

In the current version, there are 7 predefined 3x3 base rule sets:


Rule Set 0 i.e. "Standard" is the rule set used in the Seattle Robotics tutorial.  It is typically used in a FL controller where V2 (Variable 2) is a feedback variable (momentum or gradient variable). There are obviously various ways to interpret a rule set depending on the variables used.  In the Seattle Robotics tutorial V1 is a measurement in temperature (vs. a given temperature i.e. a thermostat), and V2 is the 1st order differential or gradient of such variation. The rule set is there designed to make sure the system returns to its equilibrium temperature.  In trading, we shall have very different systems so while one should always try to understand the dynamics of the control process, rule set interpretation will be vastly different.  This explains why our base models can derived to suit any sort of rule sets, from standard controllers to trading rules.

For instance, the rule set 0 looks like the following:

Standard (V1=horizontal)
V2 \ V1 Low Med High  
Low 1 3 3  



Med 1 2 3  



High 1 1 3  



One can clearly see that Var1 here supersedes Var2, i.e. if Var1 is High, the FL controller output will be High regardless of Var1.  Again, it may be recommended to read the Seattle Robotics for those who have difficulties with these concepts.

5x5 Predefined Rule Sets

Along the same line, we have a few built-in rule sets here:


Which do correspond to most possible applications, and again which can each be declined by inverting each variable.

Here is an example of a balanced rule set:

V2 \ V1 Very Low Low Medium High Very High
Very Low 3 4 4 5 5
Low 2 3 4 4 5
Medium 2 2 3 4 4
High 1 2 2 3 4
Very High 1 1 2 2 3


To better comprehend what is behind those names, and to see what's behind any of the predefined rule sets, 2  spreadsheets describe and display the different shapes:




Home Up Best viewed with MS Internet Explorer 5 and above

Page last modified: May 08, 2008
Copyright ForeTrade Technologies 21st century and thereafter