nnplot: Neural network visualization with Matplotlib

New classes are introduced that enable the plotting of neurons, any number of connexions, with weights represented by transparency. Requires matplotlib and numpy.

This was created using code originally published on StackOverflow by users Milo, OliBlum and DenisFlash at: [https://stackoverflow.com/questions/29888233/how-to-visualize-a-neural-network].

The code was intended to print neuron networks vertically, from the input layer (IL) at the bottom to the output layer (OL) at the top. It has been modified to accomodate printing text in neurons, stacking neurons vertically or horizontally and changing the colors of neurons and lines.

The code is still quite messy, but functional.

Classes

NeuralNetwork

Main class meant to be interacted with, to plot neural networks. For now, on instance initialization the number of neurons in the wides layer must be specified as the first and only necessary argument. The neuron_radius can also be specified. The drawn NN orientation can be specified with the direction keyword argument. Starting from the IL, direction = bottomtotop, toptobottom, righttoleft, lefttoright.

Then, the add_layer method can be used to create any number of layers, with the number of neurons in this new line the only necessary input. The first added layer using this method is the IL, and the last the OL. By default, the IL, hidden layers (HLs) and OL are automatically labelled, and the HLs numbered.

In the add_layer method,

• the line_color (of lines that go from the current neuron layer to the next) can be specified using a string or an numpy.array of strings of shape (current layer neuron number, next layer neuron number).
• neuron_color ibid. line-color.
• the line_weigths (of lines that go from the current layer to the next) can be specified using a float or an numpy.array of strings/floats between 0 and 1 of shape (current layer neuron number, next layer neuron number), if needed.
• neuron_weights can be specified using a float between 0 and one or a list of length equal to the number of neurons.
• neuron_text can be specified using a string a list of length of length equal to the number of neurons.

Finally, the draw method is used to plot all layers. Other plotting operations (plt.show(), title insertion, figure saving, etc.) are left to the user.

Layer

Layer object class. Not meant to be interacted with by the user.

A great number of parameters must be specified on initialization. Quite messy. Will probably be pruned somehow in the future.

Neuron

Neuron object class. Not meant to be interacted with by the user.

Specifiy its position and color on initialization. draw method used to draw it on a previously initialized figure.

Keywords: Neural Network visualization, plotting, matplotlib,