Pulsar

What Pulsar is

Pulsar monitors machine learning models in production, focusing on model drift, concept drift, and performance. It ingests model inputs and outputs into a time series database, supporting both historical and real-time data. An SDK and containerised ingestion service can be integrated into a user's machine; the current database integration is InfluxDB. Pulsar uses training data as a baseline to calculate drift and performance metrics, and its scheduler can trigger standard or customised alerts when metrics reach thresholds. A Grafana dashboard displays drift measures across time ranges and user-defined features. Listed drift measures include Kullback-Leibler divergence, Kolmogorov-Smirnov, Levene, and Wasserstein tests; performance measures include accuracy, recall, F1 score, Brier score, mean squared error, and R2 score. The features page says it is designed for large ingestion volumes without degraded performance. Pulsar is a free, open-source community edition for software engineers, data scientists, and machine learning engineers. It supports self-hosted deployment, and its homepage displays integrations including Flask, FastAPI, TensorFlow Extended, Seldon, AWS, Azure, Google Cloud, and Grafana.