def test_cmaes_random_quadratic():
# -- instantiating the function
myfct, opt_vec = get_func(n)
# -- Testing!
res = cmaes(n, myfct, stop)
assert np.abs(res - opt_vec).sum() < 1e-3
def stop(g, m, opt_vec=opt_vec, eps=eps):
assert m.ndim == 1
assert opt_vec is not None
max_g = 2e2 * m.size ** 2
if g > max_g:
return False
error = np.abs(m - opt_vec).mean()
if error > eps:
return True
else:
return False
return stop
# -- Running convergence tests with dimensionality
convergence_histories = {}
for n in N_L:
key = '%i' % n
myfct, opt_vec = get_func(n)
mystop = stop_factory(opt_vec=opt_vec)
m_opt, m_history = cmaes(n, myfct, mystop, range_min=0,
range_max=0.3*n)
convergence_histories[key] = m_history
# -- dump data
cPickle.dump(convergence_histories, open('convergence_history.pkl', 'w'))
return False
def test_cmaes_random_quadratic():
# -- instantiating the function
myfct, opt_vec = get_func(n)
# -- Testing!
res = cmaes(n, myfct, stop)
assert np.abs(res - opt_vec).sum() < 1e-3
if __name__ == "__main__":
def get_stop(max_iterations=1000):
def mystop(g):
if g <= max_iterations:
return True
else:
return False
return mystop
n = 50
max_iterations = 2e1*n**2
myfct, opt_vec = get_func(n)
stop = get_stop(max_iterations=max_iterations)
res = cmaes(n, myfct, stop, range_min=0,
range_max=0.3*n)
print("predicted num: %d" % pred)
plt.figure(1)
plt.clf()
plt.imshow(initX.reshape(28, 28), cmap=plt.cm.gray)
plt.colorbar()
plt.title("predicted num: %d" % pred)
plt.show()
mlpfunc(initX, entry=None)
#%%
from cmaes import cmaes
#%%
XOut = cmaes(fitnessfunc=mlpfunc,
N=784,
initX=initX,
maximize=True,
save=True,
savemrk='mlp4-out1',
stopsigma=1e-5)
# After see the data, I found the `sigma ` keeps going up and diverge to 1E19 after the first few trials
# so the data is not good on this sense
#%%
XOut2 = cmaes(fitnessfunc=mlpfunc,
N=784,
initX=initX,
maximize=True,
save=True,
savemrk='mlp4-out1',
stopsigma=1e-5)
#%%
