I would like to implement a GA-based function selection program for the SGI UV system (core = 512, shared memory = 4TB), which works as follows: (Function selection is the process by which the smallest subset of source functions is detected that makes discrimination between output classes is better than using the original set of functions - or in the same way - for data, for example, data consists of: {Atmospheric pressure, Temperature, myShoeSize} as independent variables, and the output is precipitation, the possible result of choosing a function would be {Atm. Pressure, Temperature}) .
GA stores a pool of parents, each of which represents a different subset of the functions. These parents must be evaluated using a machine support machine or any other machine learning method (neural network, etc.), so I want each parent to be sent to the next available cpu processor, which was evaluated using any program, and fitness was sent back to GA. Therefore, GA will be responsible for everything except the assessment (suitability) of each parent - instead, GA will send the parent to the available core and wait for the fitness result. This is where the distributed functions of the method will be (therefore I do not want different GAs to be run on different kernels, only one GA running on the same core and spawning grounds for evaluating performance in different kernels).
To take advantage of the distributed computing functions of my equipment, I would like GA to work in asynchronous mode, while there are two groups of parents, those who rated their suitability and those who are waiting for it. When there is a free core, GA takes the parent from the pool is unrated and sends it to the core. At the same time, GA takes parents from the evaluated pool, mutates them, etc. Crosses them and sends their children to an unaccounted pool, etc.
So my idea is to get the GA open source library and modify it a little with respect to the evaluation function. If the library offers this "asynchronous" mode, it will be fine. In addition to all of these, I would like the library to offer many features, for example. cellularGA. No matter what comes out, there will be an open source.
- ?
, - - "" - - ?