NAHPU is adding deep learning (DL) features, such as identifying specimens from measurements or photos. The models behind these features are trained and fine-tuned with milu, a command line tool built on the nahpu_dl crate of the NAHPU API. Basic machine learning methods, such as principal component analysis, live in the nahpu_ml crate.
Current status
Section titled “Current status”| Area | Status |
|---|---|
| Tabular models (measurements) | Work end to end |
| Image models (specimen photos) | Work end to end |
| Text models | Planned |
| ONNX import and export | Planned |
| Use inside the NAHPU app | Planned; the app does not load models yet |
In this section
Section titled “In this section”- Architecture: how milu,
nahpu_dl, and model packages fit together. - Fine-tuning: train a new model or adapt a pretrained one to your data.
- Guidelines: set up the tools, prepare data, share models, and contribute code.
Etymology
Section titled “Etymology”Like other NAHPU ecosystem tools named after deer and their close relatives, milu is named after Elaphurus davidianus, a deer native to northeastern China.