Realized EdgeList only — no simulation re-run, no kernel change (Δscience=0). Hover neurons for layer / E/I / phenotype / H / degree / position. Toggle edges by category via checkboxes + legend. Standalone HTML — no server.
N=1,000 · edges=999,000 · layers=L1, L2/3, L4, L5, L6 · cell types=E, PV, SST, VIP · phenotype distinct=18 · H ownership=no H (equilibrium 1.0, HDP disabled)
Left = what was requested in cfg.metadata (connectivity / circuit / compilation). Right = what model.params["edge_list"] materialized — the only place dynamics reads.
| Aspect | Configured spec | Realized EdgeList |
|---|---|---|
| N neurons | — | 1000 |
| Edges | spec: see configured | 999,000 (realized EdgeList) |
| Mean in-degree | — | 999 |
| Mean out-degree | — | 999 |
| Weight | 0.45 (configured within_gain) | mean 0.002798 σ 0.01033 min -0.04268 max 0.04267 |
| Delay | — | all zero (instantaneous) unique steps [0] |
| τ per edge | — | [2.0, 5.0] |
| Categories | — | E→E 430,992 · E→I 225,351 · I→E 225,351 · I→I 117,306 |
| FF / FB / local | — | FF 0 · FB 0 · local 999,000 (hierarchy ['V1']) |
| Layers → counts | { "L1": [ 0.0, 0.1 ], "L2/3": [ 0.1, 0.55 ], "L4": [ 0.55, 0.65 ], "L5": [ 0.65, 0.85 ], "L6": [ 0.85, 1.0 ] } | {"L1": 100, "L2/3": 450, "L4": 100, "L5": 200, "L6": 150} |
| Cell-type → counts | — | {"E": 657, "PV": 147, "SST": 91, "VIP": 105} phenotype distinct 18 |
{
"connectivity": {
"within_area": "all_to_all_uniform_random",
"within_gain": 0.45
},
"canonical_biophysics": true,
"layer_fractions": {
"L1": [
0.0,
0.1
],
"L2/3": [
0.1,
0.55
],
"L4": [
0.55,
0.65
],
"L5": [
0.65,
0.85
],
"L6": [
0.85,
1.0
]
},
"layer_cell_types": {
"L1": {
"E": 0.5,
"PV": 0.05,
"SST": 0.1,
"VIP": 0.35
},
"L2/3": {
"E": 0.5,
"PV": 0.25,
"SST": 0.13,
"VIP": 0.12
},
"L4": {
"E": 0.7,
"PV": 0.2,
"SST": 0.05,
"VIP": 0.05
},
"L5": {
"E": 0.85,
"PV": 0.05,
"SST": 0.05,
"VIP": 0.05
},
"L6": {
"E": 0.95,
"PV": 0.0,
"SST": 0.05,
"VIP": 0.0
}
}
}Edge sampling cap for display: 600 per category, 2400 total. Histograms sampled from up to 999000 edges with deterministic seed.
FF = cross-area low→high in hierarchy ['V1']; FB = high→low. Local = same-area (999,000 edges) not toggled separately — it is inside the E→E etc. categories. For single-area columns FF/FB are zero by construction.
Rotate/drag to inspect depth (z). Layer colors via legend — click legend entries to isolate a layer. Neuron symbols: E circle, I diamond (same hue per layer). Hover shows H, in/out-degree, phenotype, position.
Mean 0.002798 σ 0.01033 · by category: E→E 0.008906 · E→I 0.01057 · I→E -0.01057 · I→I -0.0089
All delays zero — instantaneous kernel · unique steps [0]
Mean in 999 max 999 · mean out 999 max 999
| Phenotype (layer/cell_type) distinct | 18 · {"L1/E": 50, "L1/PV": 5, "L1/SST": 10, "L1/VIP": 35, "L2/3/E": 225, "L2/3/PV": 112, "L2/3/SST": 58, "L2/3/VIP": 55, "L4/E": 70, "L4/PV": 20, "L4/SST": 5, "L4/VIP": 5, "L5/E": 170, "L5/PV": 10, "L5/SST": 10, "L5/VIP": 10, "L6/E": 142, "L6/SST": 8} |
|---|---|
| Cell-type counts | {"E": 657, "PV": 147, "SST": 91, "VIP": 105} — E/I split: E 657 / I 343 |
| Layer counts | {"L1": 100, "L2/3": 450, "L4": 100, "L5": 200, "L6": 150} |
| H ownership | {'present': False, 'enabled': False, 'mean': None, 'note': ''} — equilibrium (model.params has no hdp_initial_H) |
import jaxfne as jtfne
from jaxfne.vis.column_viewer import render_column_viewer, collect_column_viewer_data
cfg = jtfne.build_laminar_column(n=1000, ei_profile="canonical")
cfg = cfg.set_emitter("izhikevich","cortical_eig").probes(["spikes"],n_contacts=16).field(domain="laminar_column", conductivity="proxy", boundary="mean_zero_neumann")
model = jtfne.construct(cfg)
data = collect_column_viewer_data(model) # no simulation, reads realized EdgeList
render_column_viewer(model, output_path="artifacts/column_viewer_canonical_1000n.html")
# Check: data["realized"]["n_edges"] == int(model.params["edge_list"].n_edges)
# and edge_category_counts sum to n_edges.
No kernel or sampler was changed to build this viewer. The HTML is self-contained (Plotly.js via CDN) — open it in a browser, no Python server needed. Re-render with any Model (laminar, multi-area, neuronal tensor, HDP) without re-running a simulation.