How to use the dnn.Linear function in dnn

super(SigmoidLayer, self).__init__(rng, input, n_in, n_out, W, b)
        self.pre_activation = self.output
        self.output = T.nnet.sigmoid(self.pre_activation)


class ReLU(Linear):
    """ Rectified Linear Unit activation layer (max(0, W.X + b)) """
    def __init__(self, rng, input, n_in, n_out, W=None, b=None):
        if b is None:
            b = build_shared_zeros((n_out,), 'b')
        super(ReLU, self).__init__(rng, input, n_in, n_out, W, b)
        self.pre_activation = self.output
        self.output = relu_f(self.pre_activation)


class SoftPlus(Linear):
    def __init__(self, rng, input, n_in, n_out, W=None, b=None):
        if b is None:
            b_values = numpy.zeros((n_out,), dtype=theano.config.floatX)
            b = theano.shared(value=b_values, name='b', borrow=True)
        super(SoftPlus, self).__init__(rng, input, n_in, n_out, W, b)
        self.pre_activation = self.output
        self.output = softplus_f(self.pre_activation)


class BatchNormalizer(Linear):
    def __init__(self, rng, input, n_in, n_out, W=None, b=None):
        super(BatchNormalizer, self).__init__(rng, input, n_in, n_out, W, b)
        self.input = input
        x_tmp = (self.input - T.mean(self.input, axis=0)) / (T.std(self.input, axis=0) + 1.E-6)
        self.output = T.dot(x_tmp, self.W) + self.b

size=(n_in, n_out)), dtype=theano.config.floatX)
            W_values *= 4  # This works for sigmoid activated networks!
            W = theano.shared(value=W_values, name='W', borrow=True)
        if b is None:
            b = build_shared_zeros((n_out,), 'b')
        self.input = input
        self.W = W
        self.b = b
        self.params = [self.W, self.b]
        self.output = T.dot(self.input, self.W) + self.b

    def __repr__(self):
        return "Linear"


class SigmoidLayer(Linear):
    """ Sigmoid activation layer (sigmoid(W.X + b)) """
    def __init__(self, rng, input, n_in, n_out, W=None, b=None):
        super(SigmoidLayer, self).__init__(rng, input, n_in, n_out, W, b)
        self.pre_activation = self.output
        self.output = T.nnet.sigmoid(self.pre_activation)


class ReLU(Linear):
    """ Rectified Linear Unit activation layer (max(0, W.X + b)) """
    def __init__(self, rng, input, n_in, n_out, W=None, b=None):
        if b is None:
            b = build_shared_zeros((n_out,), 'b')
        super(ReLU, self).__init__(rng, input, n_in, n_out, W, b)
        self.pre_activation = self.output
        self.output = relu_f(self.pre_activation)

self.params = [self.W, self.b]
        self.output = T.dot(self.input, self.W) + self.b

    def __repr__(self):
        return "Linear"


class SigmoidLayer(Linear):
    """ Sigmoid activation layer (sigmoid(W.X + b)) """
    def __init__(self, rng, input, n_in, n_out, W=None, b=None):
        super(SigmoidLayer, self).__init__(rng, input, n_in, n_out, W, b)
        self.pre_activation = self.output
        self.output = T.nnet.sigmoid(self.pre_activation)


class ReLU(Linear):
    """ Rectified Linear Unit activation layer (max(0, W.X + b)) """
    def __init__(self, rng, input, n_in, n_out, W=None, b=None):
        if b is None:
            b = build_shared_zeros((n_out,), 'b')
        super(ReLU, self).__init__(rng, input, n_in, n_out, W, b)
        self.pre_activation = self.output
        self.output = relu_f(self.pre_activation)


class SoftPlus(Linear):
    def __init__(self, rng, input, n_in, n_out, W=None, b=None):
        if b is None:
            b_values = numpy.zeros((n_out,), dtype=theano.config.floatX)
            b = theano.shared(value=b_values, name='b', borrow=True)
        super(SoftPlus, self).__init__(rng, input, n_in, n_out, W, b)
        self.pre_activation = self.output

super(ReLU, self).__init__(rng, input, n_in, n_out, W, b)
        self.pre_activation = self.output
        self.output = relu_f(self.pre_activation)


class SoftPlus(Linear):
    def __init__(self, rng, input, n_in, n_out, W=None, b=None):
        if b is None:
            b_values = numpy.zeros((n_out,), dtype=theano.config.floatX)
            b = theano.shared(value=b_values, name='b', borrow=True)
        super(SoftPlus, self).__init__(rng, input, n_in, n_out, W, b)
        self.pre_activation = self.output
        self.output = softplus_f(self.pre_activation)


class BatchNormalizer(Linear):
    def __init__(self, rng, input, n_in, n_out, W=None, b=None):
        super(BatchNormalizer, self).__init__(rng, input, n_in, n_out, W, b)
        self.input = input
        x_tmp = (self.input - T.mean(self.input, axis=0)) / (T.std(self.input, axis=0) + 1.E-6)
        self.output = T.dot(x_tmp, self.W) + self.b


class DatasetMiniBatchIterator(object):
    """ Basic mini-batch iterator """
    def __init__(self, x, y, batch_size=BATCH_SIZE, randomize=False):
        self.x = x
        self.y = y
        self.batch_size = batch_size
        self.randomize = randomize
        from sklearn.utils import check_random_state
        self.rng = check_random_state(42)