---
license: other
pretty_name: AxoBench Population
task_categories:
  - time-series-forecasting
  - tabular-regression
tags:
  - neuroscience
  - computational-neuroscience
  - neuron-simulation
  - population-modeling
  - coreneuron
---

# AxoBench Population

AxoBench Population contains population-context neuron traces for training and
evaluating learned neuron surrogates under connected-network input
distributions. Connected AxoSim rollouts supply branch-resolved input
histories; role-correct biophysical teachers supply the target voltage and
event traces.

## Size and splits

| Split | Contexts | Traces | NPZ shards |
|---|---:|---:|---:|
| `train` | 80 | 40,000 | 480 |
| `val` | 20 | 10,000 | 120 |
| total | 100 | 50,000 | 600 |

Contexts, rather than neighboring windows from one context, define the
train/validation boundary. The corpus has exact 80/20
excitatory/inhibitory composition across six morphology conditions.

## Shard schema

Each compressed NPZ shard contains:

- `inputs`: `int8[trace, time, 1278]`;
- `targets`: `float32[trace, time, 2]`;
- `soma_mv`: `float32[trace, time]`;
- `context_index`: `int16[trace]`;
- `target_is_inhibitory`: `bool[trace]`; and
- `target_role`: excitatory or inhibitory metadata.

The native timestep is 1 ms. `train/manifest.json` and `val/manifest.json`
record shard counts, sample IDs, schema, and SHA-256 hashes.

## Validation boundary

All 600 shards passed the release validator for schema, finite values, sample
counts, role balance, manifest hashes, and disjoint context splits. This
establishes package integrity and population-compatible provenance; it does
not imply that one fixed training recipe will improve every connected
population metric.

## Loading

The Hugging Face Dataset Viewer does not natively expand these high-dimensional
NPZ tensors. Download the required shards with `hf download` or
`huggingface_hub.snapshot_download`, then load a shard with `numpy.load`.

```python
from pathlib import Path

import numpy as np

shard = next(Path("train").glob("*.npz"))
with np.load(shard, allow_pickle=False) as data:
    inputs = data["inputs"]
    targets = data["targets"]
```

## Related datasets

- `Axym-Labs/axobench`: ordinary isolated-neuron benchmark traces.
- `Axym-Labs/axobench-interventions`: paired perturbation traces.

The files are provided for research use. Users are responsible for checking
the terms of the upstream simulator, morphology, and teacher-model resources
used in their application.