Integrate Matyan into your project

Use the Matyan client in any Python script or with your ML framework. Point Run and Repo at your Matyan frontier and backend URLs.

Any Python script

Use Run(experiment=...), run["hparams"], run.track(), and run.close() as in Getting started. Set MATYAN_FRONTIER_URL and MATYAN_BACKEND_URL, or pass the backend URL to Run(repo="http://backend:53800"). The framework adapters below use the same API.

Framework integrations

Matyan provides framework adapters for popular ML libraries. Import from matyan_client:

Framework	Adapter	Usage
PyTorch Lightning	`matyan_client.pytorch_lightning.MatyanLogger`	Pass as `trainer = Trainer(..., logger=aim_logger)`. Use `repo=` with backend URL.
PyTorch Ignite	`matyan_client.pytorch_ignite.MatyanLogger`	Attach output handlers to the trainer.
Hugging Face	`matyan_client.hugging_face.MatyanCallback`	Add to `Trainer(..., callbacks=[aim_callback])`.
Keras / tf.Keras	`matyan_client.keras.MatyanCallback` or `matyan_client.tensorflow.MatyanCallback`	Add to `model.fit(..., callbacks=[MatyanCallback(...)])`.
XGBoost	`matyan_client.xgboost.MatyanCallback`	Pass to `xgboost.train(..., callbacks=[MatyanCallback(...)])`.
LightGBM	`matyan_client.lightgbm.MatyanCallback`	Pass to `lgb.train(..., callbacks=[MatyanCallback(...)])`.
CatBoost	`matyan_client.catboost.MatyanLogger`	Pass to `model.fit(..., log_cout=MatyanLogger(...))`.
Optuna	`matyan_client.optuna.MatyanCallback`	Pass to `study.optimize(..., callbacks=[aim_callback])`.
FastAI	`matyan_client.fastai.MatyanCallback`	Pass to `cnn_learner(..., cbs=MatyanCallback(...))`.
Stable-Baselines3	`matyan_client.sb3.MatyanCallback`	Pass to `model.learn(..., callback=MatyanCallback(...))`.

In all cases, use repo="http://your-backend:53800" (or the same URL via MATYAN_BACKEND_URL) so the adapter sends data to your Matyan backend and frontier.

For more detail, see the adapter source and examples in the matyan-client package (e.g. extra/matyan-client/examples/ in the repo).

What's next

While training, you can open the Matyan UI and watch runs in real time. See Manage runs and Query runs for more.