Converting APS 3.0 runs is easy: you just have to
open them with GeneXproTools 4.0 and they are automatically
converted. There are, however, a few points you must be aware of:
First, there are new parameters in GeneXproTools 4.0, such as
program size, number of literals, and used variables, that were not
available in APS 3.0. You will notice that these parameters are set
to zero in the Run Panel,
but they have all been evaluated for all the models in the run
history and are correctly shown in the History of the Report
Panel.
Second, for Classification problems GeneXproTools 4.0 shows and
keeps the classification
accuracy of all the models in the History
Panel, whereas in APS 3.0 the values of R-square
were shown instead. So, you will have to update the classification
accuracies both on the training and testing sets in the History
Panel by pressing the Refresh All button.
Third, GeneXproTools 4.0 has several new fitness
functions for Classification problems (Hits
with Penalty, Accuracy,
Squared Accuracy, SSPN,
and Correlation Coefficient),
but the old Number of Hits fitness function has now a slightly
different implementation. In the old version, there was a minimum
requirement for viability that required the number of hits to
be greater than the number of samples in the predominant class,
whereas in the new Number
of Hits there is no penalty and the fitness corresponds exactly
to the number of hits (as you have already guessed, the new Hits
with Penalty fitness function, although slightly different from
the old Number of Hits, imposes the limit for viability).
Fourth, GeneXproTools 4.0 has a total of 279 built-in
functions and comparison rules compared to just 70 in APS 3.0.
So, new naming conventions were required to accommodate all the new comparison
rules. GeneXproTools takes care of all the conversions but you
just must be aware that their representation in Karva
notation has changed.
And finally, functions Acsc, Asec, and Acot are implemented
differently in version 4.0. If any of these functions is part of
your old models you must implement it as a DDF
and then, in the Change
Seed Window, replace its symbol by the corresponding DDF.
You can use the JavaScript code below to implement the old Acsc, Asec, and Acot
functions as DDFs:
function aps3Acsc(x)
{
var varSign = 0;
if (x < 0)
varSign = -1;
else
{
if (x > 0)
varSign = 1;
else
varSign = 0;
}
return
Math.atan(x/Math.sqrt(x*x-1))+(varSign-1)*(2*Math.atan(1));
}
function aps3Asec(x)
{
var varSign = 0;
if (x < 0)
varSign = -1;
else
{
if (x > 0)
varSign = 1;
else
varSign = 0;
}
return Math.atan(x/Math.sqrt(x*x-1))+varSign*(2*Math.atan(1));
}
function aps3Acot(x)
{
return Math.atan(x)+2*Math.atan(1);
}
|
|