Try in Browser¶
The example below runs PyDSMC entirely in your browser. It uses marimo compiled to WebAssembly, so the Python code executes client-side via Pyodide.
What this demo does (and its limits)
Deep-learning frameworks such as PyTorch and Stable-Baselines3 are not available in
WebAssembly, so this demo uses a tiny hand-coded CartPole controller instead of a trained
neural agent. Everything else is the real PyDSMC: the same Evaluator, the same predefined
return property, and the same statistical confidence interval you would get locally.
Locally you would instead pass a trained agent (or any predict_fn) and typically use more
parallel environments. See Usage and Examples.
First run takes a few seconds
The first cell downloads and installs PyDSMC and its dependencies into the in-browser Python runtime. Give it a moment; subsequent runs are instant.
Live demo¶
The cell below is a live, editable notebook running entirely in your browser. On load it installs PyDSMC and statistically checks a hand-coded CartPole policy.
Edit it and run it
Click into the code, change something (e.g. epsilon for a tighter/looser confidence interval)
or the policy itself; then click Run to rerun the code and see the estimates change.
Run it locally¶
Prefer to run on a real trained agent? Install PyDSMC and follow the Examples: