Skip to content

Neural Network

Multi Layer Perceptron Model

NN

Bases: Module

Multi Layer Perceptron Model that is compatible with GPFlow

Parameters:

Name Type Description Default
dims List

List of each layer's size, needs input layer dimensions as well

 required
activation_fn str

Activation function for each layer

 'selu'
output_activation_fn str

Activation function for the last layer

 'softmax'
Source code in sgptools/kernels/neural_network.py 
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
class NN(gpflow.base.Module):
    """Multi Layer Perceptron Model that is compatible with GPFlow

    Args:
        dims (List): List of each layer's size, needs input layer dimensions as well
        activation_fn (str): Activation function for each layer
        output_activation_fn (str): Activation function for the last layer
    """
    def __init__(self, dims, 
                 activation_fn='selu', 
                 output_activation_fn='softmax'):
        super().__init__()
        self.dims = dims
        self.activation_fn = tf.keras.activations.get(activation_fn)
        self.output_activation_fn = tf.keras.activations.get(output_activation_fn)
        for i, (dim_in, dim_out) in enumerate(zip(dims[:-1], dims[1:])):
            setattr(self, 'W_{}'.format(i), tf.Variable(xavier(dim_in, dim_out),
                                                        dtype=float_type))
            setattr(self, 'b_{}'.format(i), tf.Variable(np.zeros(dim_out),
                                                        dtype=float_type))

    def __call__(self, X):
        if X is not None:
            for i in range(len(self.dims) - 2):
                W = getattr(self, 'W_{}'.format(i))
                b = getattr(self, 'b_{}'.format(i))
                X = self.activation_fn(tf.matmul(X, W) + b)
            W = getattr(self, 'W_{}'.format(i+1))
            b = getattr(self, 'b_{}'.format(i+1))
            X = self.output_activation_fn(tf.matmul(X, W) + b)
            return X