What is Neptune?
Neptune is a metadata store for MLOps that helps research and production teams better organize ML metadata. It provides a single place to organize, log, store, display, compare, and query all metadata generated during the machine learning lifecycle. The tool supports
- Experiment tracking: Organize all ML experiments in a single place.
- Model registry: Version, manage, store, and query trained models
- Monitoring ML runs: Monitor model training, evaluation, or production runs live
Neptune supports 30+ easy integrations including, Keras, TensorFlow, MLflow, PyTorch, R, and popular Python libraries.
How Does Neptune Help?
The ultimate aim of any ML experiment is to build an excellent ML model that improves iteratively. But these iterative improvements require knowledge of prior runs, model results, comparison of new ideas with the baseline, and the ability to reproduce previous work. Neptune tool helps you gain this knowledge with:
- A single place to track all ML experiments and ML model metadata
- Search and filter interface to find a specific experiment
- Tools to compare and visualize experiments
- Interface to log and display metadata for easy debugging
For production ML models, the tool supports:
- Ability to access model versions with compatible SDKs
- A place to log drifts, hardware, re-training jobs, and production models
Key Features of Neptune
Experiment and model training metadata
Neptune allows logging ML experiment and model training metadata such as metrics, hyperparameters, Console logs, predictions, hardware logs, and more.
Artifact and model metadata
Neptune allows logging artifact metadata like dataset preview, description, feature column names, paths to the dataset, and dataset size. It also supports logging dataset versions, model descriptions, model binary, and user details for a trained model.
Neptune facilitates tracking, organizing, and monitoring ML experiments. It supports advanced tracking features such as tracking metrics, diagnostics charts, and drill-down to the specific experiment.
Neptune model registry supports traceability and accessibility for ML models. It helps record datasets, parameters, code, and model binaries for every training run.
Neptune offers a visual interface to see metadata of your experiment, models, and datasets.