Add a Colab notebook for training models from log files

[cpmpercussion]

Part of #97 (option 1 from the discussion).

Adds notebooks/impsy_training.ipynb, which can be opened straight from GitHub in Colab: open in Colab. That link will work once this is merged to main.

Steps

1. %pip install impsy
2. Upload .log files or a .zip of them.
3. Count logs by dimension, taken from the -{N}d-mdrnn.log filename.
4. Choose settings with Colab form fields: model size, max epochs, patience, and dimension (0 means the most common dimension).
5. generate_dataset
6. train_mdrnn with early stopping.
7. Download the .tflite, with notes on using it in IMPSY/IMPSYpi, the AUv3 app and IMPSY Web.

The notebook only uses generate_dataset/train_mdrnn arguments that exist in the released 1.1.0, so it works with the PyPI version now and doesn't depend on #111. Outside Colab, it uses a local logs/ folder and prints the model path instead of downloading it.

The README training section and docs/get-started.md now link to the notebook.

Testing

• Ran the notebook with jupyter nbconvert --execute in a clean Python 3.12 venv with pip install impsy (1.1.0) and 6 synthetic 4d logs. Every cell ran and it trained an s model to .tflite.
• Not tested in Colab itself: the files.upload()/files.download() cells, the #@param form fields, and whether Colab asks for a session restart after the install.

Dependency fix

This also allows psutil 7 (>=6.0.0,<8, lock updated to 7.2.2). The <7 cap wasn't deliberate: it came from the original Poetry ^6.0.0 constraint, which the PEP 621 migration kept as <7. It clashed with current IPython (psutil>=7) when installing impsy in Colab. IMPSY only calls cpu_count() and virtual_memory(), which are unchanged in 7. The full test suite passes on 7.2.2.

🤖 Generated with Claude Code

https://claude.ai/code/session_016oUxTsghWZWfnCnnkKcuNv

[Copilot]

Copilot review overview

🟡 Changes recommended

The notebook has a critical Colab dependency conflict and additional validation issues remain unresolved.

Review effort: Lite
Findings: 1

Open (1)

• Resolve the psutil version conflict in the Colab setup · New

What changed in this PR

Adds a Google Colab workflow for training IMPSY models from log files and links it from the documentation.

Changes:

• Adds upload, dataset generation, training, and .tflite download steps.
• Adds configurable model and training settings.
• Links the notebook from README.md and docs/get-started.md.

File	Summary	Findings
README.md	Links to the Colab training notebook.	None.
notebooks/​impsy_training.ipynb	Implements the Colab training workflow.	Critical psutil dependency conflict (4 votes); moderate short-dataset and invalid-dimension validation issues (1 vote each).
docs/​get-started.md	Documents the Colab training option.	Nit: clarify the “no cloud component” statement (1 vote).

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[cpmpercussion]

Constraint and lock are fixed in d99cabc (psutil>=6.0.0,<8, locked at 7.2.2). The notebook installs from PyPI, so this reaches Colab with the next impsy release, which should go out soon after this merges.