2022-08-29__Visualizing_subnets.jl

# -*- coding: utf-8 -*-
# ---
# jupyter:
#   jupytext:
#     formats: ipynb,jl:light
#     text_representation:
#       extension: .jl
#       format_name: light
#       format_version: '1.5'
#       jupytext_version: 1.13.7
#   kernelspec:
#     display_name: Julia 1.7.0
#     language: julia
#     name: julia-1.7
# ---

# # 2022-08-29 • Visualizing subnets

# ## Imports

# +
#
# -

using Revise

using MyToolbox

using VoltoMapSim

# ## Params

# Based on Roxin; same as previous nb's.

d = 6
p = get_params(
    duration = 10minutes,
    p_conn = 0.04,
    g_EE = 1   / d,
    g_EI = 18  / d,
    g_IE = 36  / d,
    g_II = 31  / d,
    ext_current = Normal(-0.5 * pA/√seconds, 5 * pA/√seconds),
    E_inh = -80 * mV,
    record_v = [1, 801],
);

# ## Run sim

s = cached(sim, [p.sim]);

s = augment_simdata(s, p);

# ## Connection stats

# conn matrix is [from, to].
#
# So one row is all outputs of one neuron.
#
# Summing over cols i.e. dim two gives num outputs.

num_out = count(s.is_connected, dims=2)
num_in = vec(count(s.is_connected, dims=1));

using PyPlot

using VoltoMapSim.Plot

ydistplot("Num out" => num_out, "Num in" => num_in);

# Sure yes, 40 = 1000 * 0.04.

# Median is a bit lower than that as we post-hoc deleted the self connections.

# ## Export & draw graphs

# Let's export to Graphviz dot, to explore viz options with Gephi.

lines = ["digraph {"]
for id in s.neuron_IDs
    outputs = join(s.output_neurons[id], ", ")
    type = s.neuron_type[id]
    push!(lines, "   $(id) [type = $(type)]")
    push!(lines, "   $(id) -> {$(outputs)}")
end
push!(lines, "}")
dot = join(lines, "\n")
lines[[1:5; end-2:end]]

open(joinpath(homedir(), ".phdcache", "graph.dot"), "w") do io
    println(io, dot)
end

# Nice! Graphviz, fun.

# ![](images/gephi.png)

# All 1000 neurons are drawn (inhibitory in red). The highlighted neurons are the inputs and outputs of one neuron.

# Now I want only a subset of the net.
# Say: 1, 801, and their inputs and outputs.

lines = ["digraph {"]
for m in [1, 801]
    t = s.neuron_type[m]
    push!(lines, "   $m [type = $t]")
    for n in s.input_neurons[m]
        t = s.neuron_type[n]
        push!(lines, "   $n [type = $t]")
        push!(lines, "   $n -> $m")
    end
    for n in s.output_neurons[m]
        t = s.neuron_type[n]
        push!(lines, "   $n [type = $t]")
        push!(lines, "   $m -> $n")
    end
end
push!(lines, "}")
open(joinpath(homedir(), ".phdcache", "graph.dot"), "w") do io
    println(io, join(lines, "\n"))
end

# Seems like gephi can't handle `{…} -> ` format alas. So all inputs on sep line.

# ![](images/gephi-1-801.png)

# (Yifan Hu layout with default settings)

# So these two random neurons have 7 disynaptic connections between them:
# 2 from the exc (`1`, left) to the inh (`801`, right); and 5 from the inh to the exc. All 7 pass through an excitatory neuron.

# Same but using graphviz as layouter:

# ![](images/graphviz-1-801.svg)

# (Open in new tab to see full size).

# ## Inputs only

# Above has inputs and outputs. What if we export: inputs to both neurons, and inputs to these inputs.

# I should have a function that does this the graphviz.dot file generation..
# What's its inputs. Let's say (m => n) pairs.

function gen_dot_file(edges, s; dir=joinpath(homedir(), ".phdcache"), filename="graph.dot")
    nodes = Set()
    lines = ["digraph {"]  # DIrected graph
    for (m, n) in edges
        push!(nodes, m, n)
        push!(lines, "   $m -> $n")
    end
    for n in sort!(collect(nodes))
        t = s.neuron_type[n]
        c = (t == :exc) ? "#3fb17d" : "#ee7993"
        push!(lines, "   $n [type = $t, color = \"$c\"]")
    end
    push!(lines, "}")
    open(joinpath(dir, filename), "w") do io
        println(io, join(lines, "\n"))
    end
end;

edges = []
n = 1
for m in s.input_neurons[n]
    push!(edges, m => n)
    for l in s.input_neurons[m]
        push!(edges, l => m)
    end
end
gen_dot_file(edges, s)

# This of just the inputs to the inputs of one neuron contains already 770 neurons, i.e. almost all 1000 of them.
# Visualizing it is not very helpful, except for showing that the direct inputs are strongly interconnected (Yifan Hu layout, default gephi params):

# ![](images/gephi-in-to-in.png)

# With direct inputs to neuron `1` highlighted:

# ![](images/gephi-in-to-in-high.png)

# Most of the neurons in the tangle in the middle are neurons that connect to multiple (2 to 5) direct inputs of `1`.