Dense layers

class lasagne.layers.DenseLayer(incoming, num_units, W=lasagne.init.GlorotUniform(), b=lasagne.init.Constant(0.), nonlinearity=lasagne.nonlinearities.rectify, num_leading_axes=1, **kwargs)[source]

A fully connected layer.

Parameters:
incoming : a Layer instance or a tuple

The layer feeding into this layer, or the expected input shape

num_units : int

The number of units of the layer

W : Theano shared variable, expression, numpy array or callable

Initial value, expression or initializer for the weights. These should be a matrix with shape (num_inputs, num_units). See lasagne.utils.create_param() for more information.

b : Theano shared variable, expression, numpy array, callable or None

Initial value, expression or initializer for the biases. If set to None, the layer will have no biases. Otherwise, biases should be a 1D array with shape (num_units,). See lasagne.utils.create_param() for more information.

nonlinearity : callable or None

The nonlinearity that is applied to the layer activations. If None is provided, the layer will be linear.

num_leading_axes : int

Number of leading axes to distribute the dot product over. These axes will be kept in the output tensor, remaining axes will be collapsed and multiplied against the weight matrix. A negative number gives the (negated) number of trailing axes to involve in the dot product.

Examples

>>> from lasagne.layers import InputLayer, DenseLayer
>>> l_in = InputLayer((100, 20))
>>> l1 = DenseLayer(l_in, num_units=50)

If the input has more than two axes, by default, all trailing axes will be flattened. This is useful when a dense layer follows a convolutional layer.

>>> l_in = InputLayer((None, 10, 20, 30))
>>> DenseLayer(l_in, num_units=50).output_shape
(None, 50)

Using the num_leading_axes argument, you can specify to keep more than just the first axis. E.g., to apply the same dot product to each step of a batch of time sequences, you would want to keep the first two axes.

>>> DenseLayer(l_in, num_units=50, num_leading_axes=2).output_shape
(None, 10, 50)
>>> DenseLayer(l_in, num_units=50, num_leading_axes=-1).output_shape
(None, 10, 20, 50)
class lasagne.layers.NINLayer(incoming, num_units, untie_biases=False, W=lasagne.init.GlorotUniform(), b=lasagne.init.Constant(0.), nonlinearity=lasagne.nonlinearities.rectify, **kwargs)[source]

Network-in-network layer. Like DenseLayer, but broadcasting across all trailing dimensions beyond the 2nd. This results in a convolution operation with filter size 1 on all trailing dimensions. Any number of trailing dimensions is supported, so NINLayer can be used to implement 1D, 2D, 3D, ... convolutions.

Parameters:
incoming : a Layer instance or a tuple

The layer feeding into this layer, or the expected input shape

num_units : int

The number of units of the layer

untie_biases : bool

If false the network has a single bias vector similar to a dense layer. If true a separate bias vector is used for each trailing dimension beyond the 2nd.

W : Theano shared variable, expression, numpy array or callable

Initial value, expression or initializer for the weights. These should be a matrix with shape (num_inputs, num_units), where num_inputs is the size of the second dimension of the input. See lasagne.utils.create_param() for more information.

b : Theano shared variable, expression, numpy array, callable or None

Initial value, expression or initializer for the biases. If set to None, the layer will have no biases. Otherwise, biases should be a 1D array with shape (num_units,) for untie_biases=False, and a tensor of shape (num_units, input_shape[2], ..., input_shape[-1]) for untie_biases=True. See lasagne.utils.create_param() for more information.

nonlinearity : callable or None

The nonlinearity that is applied to the layer activations. If None is provided, the layer will be linear.

References

[1]Lin, Min, Qiang Chen, and Shuicheng Yan (2013): Network in network. arXiv preprint arXiv:1312.4400.

Examples

>>> from lasagne.layers import InputLayer, NINLayer
>>> l_in = InputLayer((100, 20, 10, 3))
>>> l1 = NINLayer(l_in, num_units=5)