ndyashas / geneticalgorithm Goto Github PK

The code goes into infinite loop when the flag "mutation" is set to 1.
Possibly due to bug in ga2/gaUtil/gaDeamon.mutateAgentObj() funtion.

Need to have Documentation

Code has to be enhanced in the following places

• ga2.gaDisc.createNextGen()

We need to have partiality in the number of children an agent gets to have per generation

Temporary files generated will stay there if not removed, which needs to be done.
By an explicit reset command ? not sure
Avoiding non-existent sessions at the start of a session ?
do we need to have this functionality ?

Storing all the variables was helping us to use less of main memory even at high number of agents in a session, but it also hits the performance hard and worsens as the number of epochs go high as ga2 will consume a lot of disk bandwidth + this is unhealthy for disks such as SSDs.

GA2 currently is hard to implement and use in an environment where the result vector is dynamic. Direct use of GA2 agent's dna as the result vector will render ga2 useless in this application, thus why not use ga2's agent dna as a structural unit for a general neural network which is used to generate dynamic output vector.

Input : AgentObj.dna , envoironmentStateVector
Output : environmentActrionVector

Description : uses a neural network which is trained by ga2 and is used by environments to get actions once an environment state is sent.

Currently , the internal random.random() function takes only integers as its input , which is not the case when spread becomes a floating point

The locking functions in ga2.gaUtil.gaDeamon are not implemented correctly and not implemented entirely
when the mode is 'UNSAFE' .

The current implementation of createNextGen uses a lot of underlying daemon functions which have hard coded parameters for heuristics, this should be brought up and centralised at only createNextGen() which has parameter intialisation.

Setting and using of the flag "genrateMode" in gaDisc needs to be implimented along with implimentation of multiparent reproduction functions.

This is due to the fact that I totally overlooked that we had resetEnv function. Will fix it in the same commit as in when the locks are fixed

ga2 will work the best when the dna vector has discrete values rather than floating point numbers.
We can add a flag to set the type of values we want to dna in ga2.gaDisc.Session