String representation for network

[sidps]

Based on Lasagne PR #324

A string representation for a Neural Network built using Lasagne. The network topology is represented using adjacency lists (currently). Listing incoming and outgoing layers for each layer in the network is optional.

Examples:
(1) A network with sequential structure: input --> dense --> dropout --> dense --> dropout --> dense

>>> print(get_network_str(output_layer))

Layer    Description                                              
-----    -----------                                              
0        <lasagne.layers.input.InputLayer object at 0x1004f1110>  
1        <lasagne.layers.dense.DenseLayer object at 0x104c86510>  
2        <lasagne.layers.noise.DropoutLayer object at 0x104c86550>
3        <lasagne.layers.dense.DenseLayer object at 0x104ca1550>  
4        <lasagne.layers.noise.DropoutLayer object at 0x104ca1590>
5        <lasagne.layers.dense.DenseLayer object at 0x104ca1810>  

(2) A network with two branches: input --> conv1, conv2 --> pool1, pool2 --> dense1, dense2 --> concat --> output

>>> print(get_network_str(output_layer, incomings=True, outgoings=True))

or

>>> layers = get_all_layers(output_layer)
>>> print(get_network_str(layers, get_network=False, incomings=True, outgoings=True))

In -->     Layer    --> Out    Description                                               
-------    -----    -------    -----------                                               
[]         0        [1, 4]     <lasagne.layers.input.InputLayer object at 0x104ca1810>   
[0]        1        [2]        <lasagne.layers.conv.Conv2DLayer object at 0x104c86550>   
[1]        2        [3]        <lasagne.layers.pool.MaxPool2DLayer object at 0x104c86510>
[2]        3        [7]        <lasagne.layers.dense.DenseLayer object at 0x104d5ec10>   
[0]        4        [5]        <lasagne.layers.conv.Conv2DLayer object at 0x104d5ec50>   
[4]        5        [6]        <lasagne.layers.pool.MaxPool2DLayer object at 0x104d5eed0>
[5]        6        [7]        <lasagne.layers.dense.DenseLayer object at 0x104d6e190>   
[3, 6]     7        [8]        <lasagne.layers.merge.ConcatLayer object at 0x104d6e1d0>  
[7]        8        []         <lasagne.layers.dense.DenseLayer object at 0x104d6e450>   

Overriding __str__ for each Layer class can change the default description of each layer.

For something prettier, there's the draw_net tool. (See Lasagne PR #174)

[ebenolson]

Great, our first utility function! Thank you.