What is this?
NNI (Neural Network Intelligence) is Microsoft's open-source AutoML toolkit that automates the machine learning lifecycle: feature engineering, neural architecture search, model compression, and hyperparameter tuning.
NNI (Neural Network Intelligence) is Microsoft's open-source AutoML toolkit that automates the machine learning lifecycle: feature engineering, neural architecture search, model compression, and hyperparameter tuning.So is it a Python library I import, or a bigger platform?It's a Python toolkit you install with pip and drive from your training code, but it can also scale experiments out to training services like Kubernetes, OpenPAI, Kubeflow, or AzureML.
It's a Python toolkit you install with pip and drive from your training code, but it can also scale experiments out to training services like Kubernetes, OpenPAI, Kubeflow, or AzureML.Which ML frameworks does it work with?Out of the box it supports PyTorch, TensorFlow, Scikit-learn, XGBoost, LightGBM, and MXNet, plus Caffe2 and more — so you plug it into the framework your model already uses.
Out of the box it supports PyTorch, TensorFlow, Scikit-learn, XGBoost, LightGBM, and MXNet, plus Caffe2 and more — so you plug it into the framework your model already uses.Is it still actively maintained?No — the repository is archived, so development has stopped. The last release was the v3.0 preview from May 2022; expect no new features or fixes.
No — the repository is archived, so development has stopped. The last release was the v3.0 preview from May 2022; expect no new features or fixes.Can I run it on Windows, or is it Linux-only?Both: its test matrix covers local training on Linux and on Windows, plus remote scenarios like Linux-to-Linux and Windows-to-Windows.
Both: its test matrix covers local training on Linux and on Windows, plus remote scenarios like Linux-to-Linux and Windows-to-Windows.What exactly can it automate for me?Four things: hyperparameter tuning with exhaustive, heuristic, and Bayesian search; neural architecture search in multi-trial and one-shot modes; model compression via pruning and quantization; plus feature engineering.
Four things: hyperparameter tuning with exhaustive, heuristic, and Bayesian search; neural architecture search in multi-trial and one-shot modes; model compression via pruning and quantization; plus feature engineering.Does it scale beyond my laptop to a cluster?Yes — Kubernetes-based training services are supported: OpenPAI, FrameworkController, and Kubeflow, plus AzureML and hybrid setups for larger tuning and NAS runs.
Yes — Kubernetes-based training services are supported: OpenPAI, FrameworkController, and Kubeflow, plus AzureML and hybrid setups for larger tuning and NAS runs.Is there video material to see it in action?Yes — there's a demo on YouTube and a Bilibili channel (last updated June 2022), and a webinar introduces Retiarii, NNI's deep learning exploratory-training framework.
Yes — there's a demo on YouTube and a Bilibili channel (last updated June 2022), and a webinar introduces Retiarii, NNI's deep learning exploratory-training framework.What are similar projects?ray, autogluon, tpot, wandb, awesome-production-machine-learning, machine-learning-zoomcamp, onnx, CNTK
ray, autogluon, tpot, wandb, awesome-production-machine-learning, machine-learning-zoomcamp, onnx, CNTKWhich topics is it in?automl, hyperparameter-tuning, nas, hyperparameter-optimization, machine-learning-algorithms, scikit-learn, distributed, distributed-training, mlops, toolkit, tensorflow, neural-network
automl, hyperparameter-tuning, nas, hyperparameter-optimization, machine-learning-algorithms, scikit-learn, distributed, distributed-training, mlops, toolkit, tensorflow, neural-network