The Streamlit LIF playground

▶ Open the live playground: localhost:8501 — start it with uv run streamlit run src/clis/streamlit_cli/app.py (details below).

Most of this lab runs batch-style: a CLI command takes a fixed set of arguments, runs a simulation, and drops a directory of artifacts (figures, metrics, a run.sh) that a notebook post then publishes. That’s the right shape for a result you want to pin down and cite.

The Streamlit app is the opposite end of that spectrum — a live, interactive surface for building intuition before you commit to a run. It re-simulates a single leaky integrate-and-fire neuron on every slider drag and redraws the voltage trace immediately.

What it does

The app (src/clis/streamlit_cli/app.py) exposes the same LIF dynamics as the neuron_cli lif CLI command, but every parameter is a slider in the sidebar:

• Input — tonic current $I$ and membrane resistance $R_m$.
• Membrane — time constant $\tau_m$, and the resting / reset / threshold potentials.
• Simulation — duration and the integration timestep $\Delta t$.

As you drag, it shows the spike count, the firing rate in Hz, and the steady-state voltage $V_\infty = V_\text{rest} + R_m I$ — flagged as spiking or subthreshold depending on whether it sits above threshold. The membrane equation it integrates is

$\tau_m \frac{dV}{dt} = -(V - V_\text{rest}) + R_m\, I,$

with a spike-and-reset rule whenever $V$ crosses $V_\text{thresh}$.

Starting the server

The app uses the project’s pinned environment, so launch it through uv (never call python/streamlit directly):

uv run streamlit run src/clis/streamlit_cli/app.py

Streamlit serves on http://localhost:8501 by default and opens a browser tab. To pick another port:

uv run streamlit run src/clis/streamlit_cli/app.py --server.port 3001

Stop it with Ctrl-C in the terminal.

How it fits the rest of the lab

The playground deliberately sits outside the CLI ↔ notebook contract: it writes no config.json / output.json / manifest.json, produces no artifacts, and isn’t bundled by any notebook runner. It’s a thinking tool, not a reproducible run. When a slider configuration produces something worth keeping, reproduce it as a pinned run with the CLI instead:

uv run python src/clis/neuron_cli/cli.py lif --current 2.5 --duration 100

That writes the artifacts a notebook post can publish.