Contents

2024-04-10 · Overview figure

Contents

2024-04-10 · Overview figure¶

include("lib/Nto1.jl")

using Revise … ✔
using Units, Nto1AdEx, ConnectionTests, ConnTestEval, MemDiskCache … ✔
using StatsBase … ✔

Sim¶

N = 6500
duration = 10minutes
spike_SNR = 40;

@time sim = Nto1AdEx.sim(N, duration);

  1.635046 seconds (21.49 k allocations: 912.268 MiB, 10.06% gc time)

sig = VI_sig(sim; spike_SNR);

include("lib/plot.jl")

import PythonCall … ✔
import PythonPlot … ✔
using Sciplotlib … ✔
using PhDPlots … ✔

plotsig(sig, [0,1000], ms, yunit=:mV);

../_images/2024-04-10__Overview-figure_7_0.png

p = 99.9
percentile(sig, p) / mV

-42.1

sigc = clip!(copy(sig), p);

plotsig(sigc, [0,1000], ms, yunit=:mV);

../_images/2024-04-10__Overview-figure_10_0.png

Aside: Otsu-like method to find spike / clipping threshold?

hist(sig[sig .> -42.1mV] / mV, xlabel="mV");

../_images/2024-04-10__Overview-figure_13_0.png

So yes it seems possible (given that our percentile is ± right..). But let’s skip here and take fixed pctile value as threshold.

Although: setting all those ceiled values to e.g. 0mV is still not good (They’re still peaks, so still lotsa noise added). Better if they’re set to e.g. the mean (around -58 mV) or somehow excluded.

Test¶

highest_firing_inputs = get_trains_to_test(sim, Nₜ = 10);

linefit_method = ConnectionTests.FitUpstroke(
    winsize = 100,  # 10 ms
    offset = 0,
)
linefit_test(sig, spiketimes) = test_conn(linefit_method, sig, spiketimes);

function spike_corr_test(post_spikes, pre_spikes)
    # weird arg order is remnant of voltage based conntests function signature:
    # `test(sig, spikes)`  ('big arg first' convention)
    
end

conntest_methods = [
    ("Spikes-only", spike_corr_test),
    ("STA height", STA_test),
    ("Line fit", linefit_test),
];

include("lib/df.jl")

using DataFrames … ✔

rows = []

for (method, f) in conntest_methods
    println(method)
    if f == spike_corr_test
        post_sig = sim.spiketimes
    else
        post_sig = sigc
    end
    t0 = time()
    spiketrain_rows = test_inputs(sim, post_sig, highest_firing_inputs, f)
    compute_time = time() - t0
    df = DataFrame(spiketrain_rows)
    sweep = sweep_threshold(df)
    row = (;
        method,
        AUC = calc_AUROCs(sweep).AUC,
        F1max = maximum(skipnan(sweep.F1)),
        compute_time,
    )
    push!(rows, row)
    println()
end

df = DataFrame(rows)

exc 100%|████████████████████████████████████████████████| Time: 0:00:04
inh 100%|████████████████████████████████████████████████| Time: 0:00:02
unc 100%|████████████████████████████████████████████████| Time: 0:00:02
exc 100%|████████████████████████████████████████████████| Time: 0:00:01
inh 100%|████████████████████████████████████████████████| Time: 0:00:01
unc 100%|████████████████████████████████████████████████| Time: 0:00:01

2×4 DataFrame

Row	method	AUC	F1max	compute_time
	String	Float64	Float64	Float64
1	STA height	0.5	0.714	9.62
2	Line fit	0.805	0.833	3.6

fig, ax = plt.subplots()
ax.bar([0, 1], df.F1max);

../_images/2024-04-10__Overview-figure_24_0.png

previous

2024-04-12 · Spikes-only connection inference

next

2024-01-11 · Re-make network hops barplots