suhmily / neat-python Goto Github PK

What steps will reproduce the problem?
1. run xor2-spiking example

What version of the product are you using? On what operating system?
svn rev 355

Please provide any additional information below.


python xor2-spiking.py 

 ****** Running generation 0 ****** 

Traceback (most recent call last):
  File "xor2-spiking.py", line 41, in <module>
    pop.epoch(200, report=True, save_best=0)
  File "/usr/local/lib/python2.6/dist-packages/neat/population.py", line
252, in epoch
    self.evaluate()
  File "xor2-spiking.py", line 22, in eval_fitness
    brain = iznn.create_phenotype(chromosome)
  File "/usr/local/lib/python2.6/dist-packages/neat/iznn/network.py", line
36, in create_phenotype
    neurons[ng.id] = Neuron(ng.bias)
NameError: global name 'Neuron' is not defined

What steps will reproduce the problem?
1. Run the Spiking XOR example
2. Note that it gets fitness 1.0 but is still returns wrong answers

I've located this to the fitness function, which uses this:
  output = brain.advance([i * 10 for i in input])
where the i variable interferes with the input index.

Solution, replace:
output = brain.advance([i * 10 for i in input])
with;
output = brain.advance([x * 10 for x in input])

Now the spiking model works for me. Woohoo!!!

This project seems not to have any documentation. If it has documentation, you 
could upload it to https://readthedocs.org/ or to PyPI (after uploading the 
package there).

What steps will reproduce the problem?
1. Execute any experiment for few generations and the memory usage
continues to increase. 

What is the expected output? What do you see instead?
Memory should be in some bound, because at one point of time we at max keep
2 generation full of ANNs.


What version of the product are you using? On what operating system?
Latest build, on Ubuntu 8.10

Please provide any additional information below.
I tried writing destructor for ANN class in /neat/nn/nn_cpp/ANN.cpp but
still the memory leaks.

Both the mutation of bias and response depend on the "prob_mutatebias":

        r = random.random
        if r() < Config.prob_mutatebias:
            self.__mutate_bias()
        if r() < Config.prob_mutatebias:
            self.__mutate_response()

In case anybody else cares about this seemingly abandoned project, attached is 
a small patch which adds an option in the config files to create minimally 
connected chromosomes as described in 
http://nn.cs.utexas.edu/keyword?whiteson:gecco05

Cheers.

What steps will reproduce the problem?
1. pip search neat -> no machine learning related results
2. https://pypi.python.org/pypi?%3Aaction=search&term=neat&submit=search -> no 
machine learning related results

Is neat-python not available via pip? Why not?