File size: 4,472 Bytes
381559e
bdcd350
 
 
 
 
 
 
381559e
 
bdcd350
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
---
pretty_name: MultiNxon2 NAS Runs (Neuraxon Game of Life)
tags:
  - neural-architecture-search
  - artificial-life
  - neuroscience
  - simulation
  - neuraxon
license: cc-by-sa-4.0
---

# MultiNxon2 — NAS Runs for the Neuraxon Game of Life

This dataset contains the raw outputs of many **Neural Architecture Search (NAS)**
runs for the **Neuraxon Game of Life** — an artificial-life simulation where
small spiking "brains" (NxErs) live, eat, mate and explore a world. Each NAS run
searches over brain hyper-parameters to find the architecture that best survives
and behaves, scored by a single `fitness` value plus a set of biology-inspired
metrics (M1–M10).

The runs were collected over several days and span Neuraxon internal versions
**v166 → v195**. Each version tweaks the search space or the fitness/selection
logic, so the collection is also a longitudinal record of how the search itself
evolved.

## Folder layout

One NAS run = one folder named by its start timestamp:

```
20260512_212222/
20260514_075838/
...
20260531_163524/
```

Each run folder contains:

| File / folder | What it is |
|---|---|
| `nas_log.csv` | **Main results table** — one row per trial (architecture). Start here. |
| `nas_best.json` | The single best architecture of the run. |
| `nas_top1.json`, `nas_top2.json`, `nas_top3.json` | The top-3 architectures. |
| `trial_NNN__arch.json` | The full hyper-parameters for trial `NNN`. |
| `trial_NNN/` | Per-trial game outputs (the actual simulation logs for that architecture). |

Inside a `trial_NNN/` folder (newer runs nest seed repeats as `rep0/`, `rep1/`,
`rep2/`; older runs use sibling `trial_NNN_s<seed>/` folders):

| File | What it is |
|---|---|
| `nxon2_<id>__BestFitness.json` | Best-of-run game state for each scoring dimension: |
| `..._BestFoodFound.json`, `..._BestFoodTaken.json` | …food found / taken, |
| `..._BestMates.json`, `..._BestTimeLived.json`, `..._BestWorldExplorer.json` | …mates, lifespan, exploration. |
| `..._KeyMetrics.txt` | Human-readable summary of the run's key metrics. |
| `..._LifespanLog.txt` | Per-agent lifespan / population log. |
| `..._MembraneDiag.txt` | Membrane / firing diagnostics. |
| `..._Completed_<timestamp>.json` | Final completed game record. |

## `nas_log.csv` columns (the important ones)

`nas_log.csv` has ~52 columns. The key ones:

- `trial_id` — index of the trial within the run.
- `fitness` — the value being optimised (higher is better).
- `arch_summary` — the **full hyper-parameter string** for that architecture
  (firing thresholds, connection probability, learning rate, topology, etc.).
- `is_global_best``1` for the run's winning architecture.
- `M1``M10` — paper-fidelity metrics (e.g. `M1` = excitatory firing band,
  `M1_neutral` / `M1_inh` = rest / inhibition fractions, plus synchrony,
  plasticity and other dynamics measures).
- Survival / behaviour: `final_alive`, `peak_alive`, `alive_mean`,
  `went_extinct`, `surv_score`, `expl_rate`.
- Reliability: `n_repeats`, `n_repeats_ok`, `fitness_std_reps`, `M1_std`
  (later versions re-run the same architecture across seeds to average out noise).
- Bookkeeping: `completed_at_iso`, `wall_actual_s`, `total_rounds`, `error`.

## Quick start

```python
import pandas as pd

# point at any run folder
df = pd.read_csv("20260531_163524/nas_log.csv")

# best architecture in that run
best = df.sort_values("fitness", ascending=False).iloc[0]
print(best["fitness"], best["arch_summary"])
```

## Notes

- All paths in the original logs use `\...`; that prefix is just the
  source drive and can be ignored.
- This is **raw research output**: column sets and folder conventions shift
  slightly between versions as the search evolved.

## License

Released under CC-BY-4.0-SA

## Citation
```bibtex
@dataset{NeuraxonGameOFLifeResearhNAS-5-MultiNxon2NAS
  title={Neuraxon Game of Life 5 Research Dataset: NAS Multi Nxon Exploration},
  author={Vivancos, David and Sanchez, Jose},
  year={2026},
  publisher={Hugging Face},
  version={1.0.0},
  url={https://huggingface.co/datasets/DavidVivancos/MultiNxon2NAS}
}
```
## Authors & Curators
*   **David Vivancos** / Artificiology Research (https://artificiology.com) - [Qubic Science](https://qubic.org/)
*   **Dr. Jose Sanchez** / UNIR - [Qubic Science](https://qubic.org/)
**Contact:** For questions or issues, please open a GitHub issue at [https://github.com/DavidVivancos/Neuraxon](https://github.com/DavidVivancos/Neuraxon).