spikecraft
Write a spiking neuron as equations, get a trainable PyTorch layer. Or pick one from the zoo — every zoo model is itself just an equation preset you can open and edit.
pip install spikecraft # after publishing; for now: pip install -e .
1. Your own neuron from equations
import spikecraft as sf
neuron = sf.Neuron(
equations="""
dv/dt = (-(v - v_rest) + R*I) / tau
da/dt = -a / tau_a
""",
threshold="v > v_th + beta*a",
reset="v = v_reset; a += 1",
params=dict(tau=10.0, tau_a=100.0, beta=0.2, R=1.0, v_rest=0.0, v_reset=0.0, v_th=1.0),
learnable=["tau", "beta"], per_neuron=True, # trained by backprop, one value per neuron
surrogate="atan", # fast_sigmoid | sigmoid | atan | triangle | gaussian
solver="exp_euler", # exact exponential step where the ODE allows
)
layer = sf.SpikingLinear(neuron, in_features=128, out_features=64)
spikes = layer(x) # x: [T, B, 128] -> [T, B, 64]
Iis the input current (rename withinput="...").dtand time constants share one unit (ms by convention).- Functions:
exp, log, sqrt, tanh, sin, cos, abs, max, min, sigmoid, heaviside. - Reset supports
=,+=,-=,*=. Leavereset=Noneand usethreshold_mode="rising"for models without reset (e.g. Hodgkin–Huxley). - Mistakes are caught early: unknown symbols, malformed equations, resets on missing states.
2. Model zoo
from spikecraft import zoo
zoo.list_models()
# ['ALIF', 'AdEx', 'EIF', 'HodgkinHuxley', 'Izhikevich', 'LIF', 'QIF',
# 'ResonateAndFire', 'SynapticLIF', 'TwoCompartment']
lif = zoo.LIF(tau=20.0, learnable=["tau"])
izh = zoo.Izhikevich(firing="chattering") # regular_spiking, intrinsically_bursting, fast_spiking, ...
hh = zoo.HodgkinHuxley()
print(zoo.AdEx()) # shows its equations and params
Edit a preset's equations directly:
spec = zoo.spec("LIF")
spec["equations"] = "dv/dt = (-(v - v_rest) + R*I + bias) / tau"
spec["params"]["bias"] = 0.1
my_lif = sf.Neuron(**spec)
| Model | States | Notes |
|---|---|---|
| LIF | v | leaky integrate-and-fire |
| SynapticLIF | i, v | current-based synapse filter |
| ALIF | v, a | adaptive threshold |
| QIF / EIF | v | quadratic / exponential IF |
| Izhikevich | v, u | 6 named firing types |
| AdEx | v, w | Brette–Gerstner, pA / mV |
| HodgkinHuxley | v, m, h, n | gates integrated exactly (Rush–Larsen) |
| ResonateAndFire | x, y | damped oscillator |
| TwoCompartment | vd, vs | dendrite input, somatic spike |
3. Layers and utilities
SpikingLinear,SpikingConv2d, andSpiking(any_module, neuron)for any synapse.recurrent=Trueadds recurrent weights.- Neurons are plain
nn.Modules:neuron.step(I, state)for custom loops,neuron(I_seq, return_states=True)for traces. sf.encoding.rate,latency,repeat,spike_count.torch.save/state_dictwork as usual.
How it works
Equations are parsed with SymPy and compiled to PyTorch. For each state x the solver computes A = ∂f/∂x; if A doesn't depend on x, the step is the exact exponential update x ← x·e^{AΔt} + (e^{AΔt}−1)/A · (f − Ax), otherwise forward Euler. Spikes use a Heaviside forward pass with a surrogate gradient backward pass.
Development
pip install -e ".[dev]"
pytest
python examples/train_custom_vs_zoo.py
python -m build && twine upload dist/*
License
MIT
Metadata
Release files for spikecraft 0.1.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| spikecraft-0.1.0.tar.gz | 14.4 kB | Without provenance | Details for spikecraft-0.1.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| spikecraft-0.1.0-py3-none-any.whl | Python 3 | none | any | Details for spikecraft-0.1.0-py3-none-any.whl |
Total release size: 30.5 kB
Release files / spikecraft-0.1.0.tar.gz
| Download URL | spikecraft-0.1.0.tar.gz |
|---|---|
| Size | 14.4 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
304a0476dba25b36909f3b8bfb63497d1dc5a5365c88ac96c8ee921d3ee0ea4e
|
|
BLAKE2b-256 checksum How to use checksums |
f6006a3a37173168b6f58fbc24175214cf9faead1ca4c499f0c2bc667ab04896
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.11.18
|
Release files / spikecraft-0.1.0-py3-none-any.whl
| Download URL | spikecraft-0.1.0-py3-none-any.whl |
|---|---|
| Size | 16.0 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
ace77d60850030b6fa9a7d4bd5d257371a5f831676f9e6454d3b2d0327868780
|
|
BLAKE2b-256 checksum How to use checksums |
7bbc85dbb6eece0ca369dac4ba68b3437329d9ce208e6fe1c4bac7c292191f95
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.11.18
|
Security for spikecraft 0.1.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.