def __init__(self, rngs, input_layer, Lshape, traits, activation):
super(ConvLayer, self).__init__(input_layer, traits, "Conv")
self.rng = rngs[0]
self.l2decay = traits['l2decay']
filter_shape = Lshape[1]
# The number of input channels must match number of filter channels
assert Lshape[0][1] == filter_shape[1]
self.pad = traits['padding']
self.W = NNl.gen_weights(self.rng, filter_shape, 0, traits['initW'])
# convolve input feature maps with filters
# Using Alex K.'s fast CUDA conv, courtesy of S. Dieleman
self.x = self.input_layer.output(False)
conv_op = FilterActs(pad=self.pad, partial_sum=1)
input_shuffled = (self.x).dimshuffle(1, 2, 3, 0) # bc01 to c01b
filters_shuffled = (self.W).dimshuffle(1, 2, 3, 0) # bc01 to c01b
contiguous_input = gpu_contiguous(input_shuffled)
contiguous_filters = gpu_contiguous(filters_shuffled)
out_shuffled = conv_op(contiguous_input, contiguous_filters)
self.conv_out = out_shuffled.dimshuffle(3, 0, 1, 2) # c01b to bc01
# store parameters of this layer
self.params = [self.W]
def __init__(self, rngs, input_layer, Lshape, traits, activation):
super(OutputLayer, self).__init__(input_layer, traits, "Output")
self.out_shape = (Lshape[0], Lshape[1])
self.W_shape = Lshape[1:]
self.activation = activation
self.l2decay = traits['l2decay']
if len(Lshape) != 3:
print("Logistic regression shape must be (2,), it is,", Lshape)
# Initialize weights and biases (can load values later)
self.W = NNl.gen_weights(rngs[0], self.W_shape, 0, traits['initW'])
self.b = Tsh(np.zeros((Lshape[2],), dtype=Tfloat))
self.params = [self.W, self.b]
def __init__(self, rngs, input_layer, Lshape, traits, activation):
super(FCLayer, self).__init__(input_layer, traits, "FC")
self.p_retain = (1. - traits['dropout'])
self.rng = rngs[0]
self.srng = rngs[1]
self.out_shape = (Lshape[0], Lshape[2])
self.W_shape = Lshape[1:]
self.activation = activation
self.l2decay = traits['l2decay']
self.d_rec = input_layer.output(False)
self.best_error = np.inf
if len(Lshape) != 3:
print "FC layer shape must be (2,), it is,", Lshape
self.W = NNl.gen_weights(self.rng, self.W_shape, 0, traits['initW'])
self.b = Tsh(np.zeros(Lshape[2], dtype=Tfloat))
self.ib = Tsh(np.zeros(Lshape[1], dtype=Tfloat))
self.params = [self.W, self.b,]
self.pt_params = [self.W, self.b, self.ib]
