Tutorials¶
Nine notebooks, one per main feature of Penicillium. Each one runs end to end in a few minutes and is shown here
with its output. To run them yourself, download them with the button at the top of each page, or open them from
docs/tutorials/ in a clone of the repository, after installing the notebook extra:
pip install -e ".[notebooks]" # tutorial 8 also needs: pip install -e ".[jax]" diffrax
| # | Tutorial | What you learn |
|---|---|---|
| 1 | Your first batch | Run a batch, read the results, batch phases, every channel by group, compare seeds, save the data |
| 2 | Recipes and control | Read and change the recipe, compare recipes on the same batches, step through a batch with your own control rules, control PAA from lab results |
| 3 | Options, faults and disturbances | SimulationConfig, disturbances on and off, the IndPenSim faults and their labels, setpoints, batch length, initial conditions, replaying a batch |
| 4 | Lab data, Raman spectra and measurement noise | Delayed lab assays, Raman spectra and a 3D waterfall, realistic sensor and assay errors, one file with everything |
| 5 | Campaigns and fault detection | One-call data sets; an MPCA monitor built in three stages (training, validation, test) and scored per batch and per sample |
| 6 | A Raman soft sensor | Calibrate a PLS model on spectra and lab results, test it on new batches |
| 7 | Reinforcement learning | The Gymnasium environments, training a policy with the cross-entropy method and testing it on unseen batches, Minari datasets and behaviour cloning |
| 8 | The differentiable model (JAX) | Exact Jacobians, parameter sensitivity, gradients through integration |
| 9 | The command line | Generate data sets without Python, start the operator console |
The tutorials link to the matching pages of the user guide, which explain each feature in full.
Longer case studies (replications of Goldrick et al. 2015 and 2019, a statistical check of the noise model) are
in the repository's notebooks/ folder.