- Tune the hyperparams
- Test different network types
- Compare different input datasets
How to use ANN model as fitness function in genetic algorithm
26 views (last 30 days)
I am using ANN for modeling the reactive extraction efficiency of gallic acid. Now I want to optimize this model using GA, I am using the calling function 'optimtool', but after entering the fitness function and number of variables and running the silver, it shows 'Parameters.hiddenSizes contains negative value'.... I am not very proficient at all these, I have just started learning.. So somebody please help me in first identifying the correct fitness function to be called and secondly hobw to deal with this error, also in another problem, the error was "undefined function 'train' for input arguments of type double"... I am getting anything about this... Kindly help me out, I have little time left to complete it, any help would be appreciated
David Willingham on 18 Jun 2021
Can I ask what in the model are you trying to optimize?
Typically Optimization is used to determine the optimal parameters of some fundamental equation that have a solution space that may not be completely bounded.
On the other hand ANN's are used to understand a solution space bounded by the historical data they are trained on.
It's for this reason ANN's, another technique called "Experimentation" is commonly used. This is where you setup an experiments which will run "trials" to tune, test or compare the network. For example:
MATLAB's Experiment Manager App helps you setup and manage these types of experiments. It allows you to setup the range of the parameters you wish to sweep over or use Baysian Optimization to fine tune them.
Here are 2 videos that show how it works: