PipelineAI Platform

From GM-RKB
(Redirected from PipelineAI)
Jump to navigation Jump to search

A PipelineAI Platform is an enterprise end-to-end ML framework.



References

2017

  • http://pipeline.ai/features/
    • Fast Model Deployments / Deploy Models to Production in 1-Click
    • Safe Experiments in Prod / Test on 1% of Live and Shadow Traffic
    • 360º Model Comparisons / Compare Real-Time and Batch Metrics
    • Automatic Traffic Shifting / Shift Traffic to Winning Model or Cloud
    • Model + Runtime Tuning / Tune Both Model and Runtime Params
    • Deep Prediction Tuning / Monitor Predictions at All Layers
    • Live Prediction Streams / View Live Predictions in Real-Time
    • Real-Time Model Updates / Improve Models with Crowd-Sourcing
    • Auto-Optimized Models / Optimize Models for Fast Predictions
    • Secure Private Hosting / Host in Your Own Cloud or Data Center

2017b

  • https://www.crunchbase.com/organization/pipelineio
    • QUOTE: PipelineAI: End-to-End ML and AI Platform for Real-time Spark and Tensorflow Data Pipelines

       PipelineAI’s primary focus is Continuous ML and AI Model Deploying, Optimizing and Scaling. Using a combination of open source and proprietary technologies, the PipelineIO service enables data scientists to rapidly train, test, optimize, deploy, and scale models in production directly from a Jupyter Notebook or command-line interface.

      We treat model optimizing and serving as a first-class citizen in the modern data pipeline - alongside model training. We give data scientists and engineers the freedom to quickly deploy, test, and rollback (if needed) their models directly in production. A concept we practiced heavily at Netflix, this freedom comes with responsibility. PipelineIO provides the tooling, infrastructure, and dashboards necessary to responsibily manage production directly - and with no downtime.

      We currently support models built with Spark, Tensorflow, Scikit-learn, XGBoost, and R. We are constantly tuning and optimizing our runtime to provide the best price per prediction available - even across multiple data centers and cloud vendors. Our hybrid-cloud “auto-shift” technology compliments the traditional, single-cloud “auto-scale” technology.

      PipelineAI opens up new ways to increase performance of your predictions, improve the uptime of your model serving infrastructure, and reduce cost per prediction.