🧭 Neptune.ai Flashcards
Track, Visualize, and Organize ML Experiments at Scale
🔍 What is Neptune.ai?
Neptune is a metadata store for MLOps. It logs, stores, and visualizes ML experiments and models.
📦 Installation
Install with pip install neptune or neptune-client. Requires a project API token.
🔐 API Key
Authenticate using an API token and workspace/project name via environment variables or config files.
🧪 Init Run
Use neptune.init_run() to start a new experiment run and log metadata.
📈 Logging Metrics
Log metrics using run["accuracy"] = 0.92 or run["loss"].log(loss_value).
📋 Logging Parameters
Pass a dictionary of parameters like run["params"] = {"lr": 0.01, "batch": 64}.
📂 Model Registry
Use Neptune as a lightweight model registry to store and compare production-ready models.
🧮 Tracking Artifacts
Upload files with run["model"].upload("model.pkl") to save artifacts like weights or logs.
🎯 Projects & Tags
Organize runs using project structure and tags for filtering, comparison, and traceability.
🔄 Autologging
Supports autologging with Keras, XGBoost, LightGBM, CatBoost, PyTorch Lightning, and more.
📊 Web UI
Use the Neptune dashboard to explore experiments, view plots, track lineage, and compare runs.
🛠️ Use Cases
Ideal for ML teams doing reproducible research, MLOps, model debugging, and collaboration.