Developing better products faster
| |
|||
|
|
INForm Enhancements
The latest release (September 2007) of INForm brings a number of new features, especially improvements in optimization. These supplement the things we introduced in the last release - parallel coordinate plots (so that you can examine the optimum solutions) and a bigger range of mathematical functions (so that you can use mathematical, as well as neural network, models in the optimization) - to give an exceptionally powerful optimization tool.
In this new version, you can define 'sets' of related ingredient types. The set of ingredients may combine together in different configurations to produce a specific value.
Suppose, for example, that you have 6 possible additives, but want to include only two of them in your formulation - but you don't know which two would give the best results.
In earlier versions of INForm you could define a constraint script to be A + B + C + D + E + F = 100. But if you did, all 6 ingredients were likely to be assigned non-zero values, unless you fixed some of them at zero. There was no way of restricting the sum to have a value of 100 just by combining any 2 of the 6 possible ingredient values. What you actually want is for the program to try the various combinations for you, automatically.
In the new INForm , by defining the ingredients as being members of a 'set', you can specify how many members of the set should be used, and the total value to which they sum. The optimization then tells you the best combination to use, subject to the limitations that you put on the number of additives you want.
Multiple optima found automatically
For many formulation optimizations, there might be more than one solution to the problem. In common with other optimization procedures, INForm's original genetic algorithms found just the one 'best' optimum. In some cases, though, this could depend on the starting point that you selected.
In INForm v3.6, you can now select multiple populations for the optimization. Each population is optimized by the genetic algorithms. There is no exchange of genetic information between the populations, so each evolves independently. Therefore, each population might give a different solution to the optimization, and the best of these (the global minimum) is displayed on the Optimizer.
More control over optimization parameters
We've described above how you can use more than one population, evolving each separately to a different solution. We've also allowed more control of the optimization parameters. You can now change the population size (previously fixed at 100) as well as the random seed.
Best Match calculation improved
Missing data are now ignored in the 'best match' calculation. So, if you have a missing value (specified by -99999) for one of the variables in a data record, that variable will be ignored in the Euclidean Distance calculation for that data record. Before, 'missing data' meant that the Best Match calculation was unfairly penalized, giving a very poor match for any record where there was a missing value.
Previous versions of INForm are incompatible with Windows Vista - however, this version has been modified to work on Vista, future-proofing your work.
We've also upgraded the software protection system to the latest version.
This document maintained by
webmaster@intelligensys.co.uk.
Copyright © 2006 Intelligensys Ltd