Draft plotting function for network API

[michaeldeistler]

closes #179

@kyralianaka @jnsbck plotting is on the way.

net = jx.Network(cells, conns)
net.vis()

For feedforward NNs:

net = jx.Network(cells, conns)
net.vis(layers=[5, 10, 1])  # 5 input neurons, 10 hidden, 1 output.

[jnsbck]

Cool! Should this be recursive? I.e. vis/plot/render also being callable on a cell / branch level and being called by Network.vis?

[jnsbck]

Some misc. ideas / considerations from my side:

• If I am not mistaken, connections are only tracked via the parents attr. I am guessing morphologies / graphs are reconstructed from this, which means 3D structure is lost, that we get from .swc files. It would be nice to somehow store the x,y,z,d (d=diameter) coords in a position attr or sth of each compartment. Then we could plot the 3D morphology or projections thereof. Then we could add a kwarg project="xy" or dims="xy" or sth. similar to choose on which plane to project the neuron and even render it in 3d? (maybe longterm), which could be added as option to `dims="xyz".
• should we build from scratch or parse graph and use some existing graph plotting lib?
• I really like the detail kwarg. Could be useful on a cell level as well. I.e. plot thickness of compartments or just the raw graph.
• Maybe MorphoPy's draw_2D and draw_3D could be a starting point / provide some inspiration.

Will also continue to think some more about this.

[michaeldeistler]

Great suggestsions, I agree with almost all points. Especially with keeping (x,y,z). I would prefer using graph libraries only for plotting. Otherwise I am worried we will run into autograd issues in the long run.

[michaeldeistler]

Merging now to unblock Kyra, but I think this can definitely be improved.

[jnsbck]

Cool! One addition I would make is, nx.draw takes ax as a kwarg. I think it would be convenient for plotting if vis would take an ax kwarg as well.
Furthermore, draw takes a lot of kwargs (see networkx.draw_networkx()). It would be nice to link it in the docs.
We could also provide some nice defaults, i.e. edge-thickness / edge-color $\propto$ Weights or arrows in direction of synapse.