time_step = 2
time_window = [-50, 20]
T = 2000
simulation = np.load('linli_generalization.npy')
spike_events = simulation[:-1]
spike_events[0] = np.floor(spike_events[0])
target = simulation[-1]
target = target[range(0,target.shape[0],time_step)]
input = ts.nest_2_input(spike_events, 3*T, time_window=time_window, \
time_step=time_step)
#sklms = SpikeKLMS(kernel="pop_mci", growing_criterion='novelty', growing_param=[2., 10.], ksize=5,learning_rate=.0005, n_jobs=4)
sklms = SpikeKLMS(kernel="pop_mci", growing_criterion='quantized', \
growing_param=[0], ksize=5, learning_rate=.0005, \
loss_function='least_squares', correntropy_sigma=1., n_jobs=1, \
gamma=.01, dropout=0)
print sklms
for i in range(3):
sklms.fit(input[:T/2], target[:T/2])
sklms.X_transformed_ = sklms.transform(input)
G = gridspec.GridSpec(3,1)
pl.figure()
pl.subplot(G[:-1])
pl.plot(spike_events[0], spike_events[1], '.')
pl.subplot(G[-1])
pl.plot(target)
pl.plot(sklms.X_transformed_)
pl.show()
import pylab as pl
from sklearn.preprocessing import scale
import spike_klms2
pl.close('all')
reload(spike_klms2)
from spike_klms2 import SpikeKLMS
from scipy.io import loadmat
sts = loadmat("sts.mat").get("sts")
X = range(len(sts))
for i in xrange(len(X)):
X[i] = squeeze(sts[i,0])
if not(X[i].shape):
X[i] = X[i][newaxis]
d = squeeze(loadmat("targets.mat").get("targets"))
t = linspace(1,500,500)
sklms = SpikeKLMS(kernel="mci", gamma=.05, learning_rate=.0003, ksize=.005, growing_criterion="dense",\
growing_param = [15, .01], dropout=5)
sklms.fit_transform(X, d)
pl.plot(sklms.X_transformed_)
pl.plot(d)
pl.figure()
pl.plot(sklms.coeff_)
pl.show()
fe = np.mean((d - sklms.X_transformed_)**2)/np.var(d)
print "Final error: %f" % fe
time_step = 2
time_window = [-50, 20]
T = 2000
simulation = np.load('linli_generalization.npy')
spike_events = simulation[:-1]
spike_events[0] = np.floor(spike_events[0])
target = simulation[-1]
target = target[range(0, target.shape[0], time_step)]
input = ts.nest_2_input(spike_events, 3*T, time_window=time_window, \
time_step=time_step)
#sklms = SpikeKLMS(kernel="pop_mci", growing_criterion='novelty', growing_param=[2., 10.], ksize=5,learning_rate=.0005, n_jobs=4)
sklms = SpikeKLMS(kernel="pop_mci", growing_criterion='quantized', \
growing_param=[0], ksize=5, learning_rate=.0005, \
loss_function='least_squares', correntropy_sigma=1., n_jobs=1, \
gamma=.01, dropout=0)
print sklms
for i in range(3):
sklms.fit(input[:T / 2], target[:T / 2])
sklms.X_transformed_ = sklms.transform(input)
G = gridspec.GridSpec(3, 1)
pl.figure()
pl.subplot(G[:-1])
pl.plot(spike_events[0], spike_events[1], '.')
pl.subplot(G[-1])
pl.plot(target)
pl.plot(sklms.X_transformed_)
pl.show()
