def __init__(self,
num_input,
num_hidden,
input_layers=None,
name="",
go_backwards=False):
"""
MGU Layer from 周志华paper
Takes as input sequence of inputs, returns sequence of outputs
"""
self.name = name
self.num_input = num_input
self.num_hidden = num_hidden
if len(input_layers) >= 2:
# axis=1 an row xiang jia
self.X = T.concatenate(
[input_layer.output() for input_layer in input_layers], axis=1)
else:
self.X = input_layers[0].output()
self.h0 = theano.shared(floatX(np.zeros(num_hidden)))
self.s0 = theano.shared(floatX(np.zeros(num_hidden)))
self.go_backwards = go_backwards
self.W_fx = random_weights((num_input, num_hidden),
name=self.name + "W_fx")
self.W_fh = random_weights((num_hidden, num_hidden),
name=self.name + "W_fh")
self.U_h = random_weights((num_input, num_hidden),
name=self.name + "U_h")
self.W_h = random_weights((num_hidden, num_hidden),
name=self.name + "W_h")
self.b_f = zeros(num_hidden, name=self.name + "b_f")
self.b_h = zeros(num_hidden, name=self.name + "b_h")
self.params = [self.W_fx, self.W_fh, self.b_f]
self.output()
def __init__(self,
num_input,
num_hidden,
input_layers=None,
name="",
go_backwards=False):
self.name = name
self.num_input = num_input
self.num_hidden = num_hidden
if len(input_layers) >= 2:
self.X = T.concatenate(
[input_layer.output() for input_layer in input_layers], axis=1)
else:
self.X = input_layers[0].output()
self.h0 = theano.shared(floatX(np.zeros(num_hidden)))
self.s0 = theano.shared(floatX(np.zeros(num_hidden)))
self.go_backwards = go_backwards
self.W_gx = random_weights((num_input, num_hidden),
name=self.name + "W_gx")
self.W_ix = random_weights((num_input, num_hidden),
name=self.name + "W_ix")
self.W_fx = random_weights((num_input, num_hidden),
name=self.name + "W_fx")
self.W_ox = random_weights((num_input, num_hidden),
name=self.name + "W_ox")
self.W_gh = random_weights((num_hidden, num_hidden),
name=self.name + "W_gh")
self.W_ih = random_weights((num_hidden, num_hidden),
name=self.name + "W_ih")
self.W_fh = random_weights((num_hidden, num_hidden),
name=self.name + "W_fh")
self.W_oh = random_weights((num_hidden, num_hidden),
name=self.name + "W_oh")
self.b_g = zeros(num_hidden, name=self.name + "b_g")
self.b_i = zeros(num_hidden, name=self.name + "b_i")
self.b_f = zeros(num_hidden, name=self.name + "b_f")
self.b_o = zeros(num_hidden, name=self.name + "b_o")
self.params = [
self.W_gx,
self.W_ix,
self.W_ox,
self.W_fx,
self.W_gh,
self.W_ih,
self.W_oh,
self.W_fh,
self.b_g,
self.b_i,
self.b_f,
self.b_o,
]
self.output()
def reset_state(self):
self.h0 = theano.shared(floatX(np.zeros(self.num_hidden)))
self.s0 = theano.shared(floatX(np.zeros(self.num_hidden)))