def main():
lis = pimadataf.give_datainshared()
rest_setx, rest_sety = lis[0] #tuple of two shared variable of array
test_setx, test_sety = lis[1] #tuple of shared variable of array
y = T.ivector('y')
newmlp = MLP(n_in, n_out, n_hid, rest_setx, rest_sety, test_setx,
test_sety)
#error=0.5*T.mean((y-output.reshape((x.shape[0],)))**2)
#finalerror=T.mean(abs(y-(output.reshape((x.shape[0],)))))
fun = theano.function([], rest_sety)
fund = theano.function([], test_sety)
params = newmlp.params
cost = newmlp.cost_func(y)
learning_rate = 0.01
gparams = [T.grad(cost, j) for j in params]
updates = [(param, param - learning_rate * gparam)
for param, gparam in zip(params, gparams)]
train_model = theano.function([],
cost,
updates=updates,
givens={y: rest_sety})
#fun1=theano.function([x],output.reshape((x.shape[0],)))
test_model = theano.function([],
newmlp.find_error(y),
givens={y: test_sety})
print(fun())
epochs = 1000
for i in range(1, epochs):
#p=train_model(rest_setx.get_value(),fun())
p = train_model()
if i % 100 == 0:
st = newmlp.turn_weights_into_chromosome()
newmlp.set_weights_from_chromosome(st)
print("been through here")
#print(fun1(rest_setx.get_value()))
#print(fun()-fun1(rest_setx.get_value()))
#heyout=fun1(rest_setx.get_value())
#print(np.where(heyout>0.5,1,0))
#print(w1.get_value(),w2.get_value())
#print("mid",printmidout(rest_setx.get_value()))
#print("out",printout(rest_setx.get_value()))
print(p)
print("here testing", test_model())
def __init__(self,rng, max_hidden_units, size=50, limittup=(-1,1)):
self.dimtup = pimadataf.get_dimension()
rest_set, test_set = pimadataf.give_data()
tup = pimadataf.give_datainshared()
self.rng=rng
self.size = size
self.max_hidden_units = max_hidden_units
self.list_chromo = self.aux_pop(size, limittup) #a numpy array
self.fits_pops = []
self.trainx = rest_set[0]
self.trainy = rest_set[1]
self.testx = test_set[0]
self.testy = test_set[1]
self.strainx, self.strainy = tup[0]
self.stestx, self.stesty = tup[1]
self.net_err = Network.Neterr(inputdim=self.dimtup[0], outputdim=self.dimtup[1], arr_of_net=self.list_chromo, trainx=self.trainx, trainy=self.trainy, testx=self.testx, testy=self.testy,strainx=self.strainx, strainy=self.strainy, stestx=self.stestx, stesty=self.stesty)
self.net_dict={} #dictionary of networks for back-propagation, one for each n_hid
high=np.sqrt(6. / (n_in + n_hid)),
size=(n_in, n_hid)),
dtype=theano.config.floatX)
w1 = theano.shared(value=W_values, name='w1', borrow=True)
b_values = np.zeros((n_hid, ), dtype=theano.config.floatX)
b1 = theano.shared(value=b_values, name='b1', borrow=True)
w2 = theano.shared(value=np.zeros((n_hid, n_out), dtype=theano.config.floatX),
name='w2',
borrow=True)
b2 = theano.shared(value=np.zeros((n_out, ), dtype=theano.config.floatX),
name='b2',
borrow=True)
lis = pimadataf.give_datainshared()
rest_setx, rest_sety = lis[0] #tuple of two shared variable of array
test_setx, test_sety = lis[1] #tuple of shared variable of array
x = T.matrix('x')
y = T.ivector('y')
lin_midout = T.dot(x, w1) + b1
midout = T.tanh(lin_midout)
params1 = [w1, b1]
printmidout = theano.function([x], midout[0])
lin_out = T.dot(midout, w2) + b2
output = T.nnet.sigmoid(lin_out)
printout = theano.function([x], output[0])
params2 = [w2, b2]