Home

This is a living document containing NNI team's current priorities as well as release notes for previous releases. Future roadmap last updated 11/07/2018.

NNI Roadmap The following is a summary of the NNI team's backlog for the next 6 months. Some completed items are included to provide the context and progress of the work. If you have any questions or suggestions about this roadmap, you are highly encouraged to submit a github issue directly to nni project.

OS and installation supports

• Support Linux
  • Support pip install
  • Support source codes install
  • Support package install
• Support Mac
• Support Windows

General

• Support Python API for user to define tuners and assessors
• Support Python API for user to wrap trial code as NNI deployable codes
• cmdline tool NNICTL for experiment management

Built-in Algorithms (Tuner and Assessors)

• Support hyperopt_tpe (TPE)
• Support hyperopt_annealing
• Support hyperopt_random
• Support evolution_tuner
• Support medianstop algorithm for early stop assessor
• Support automatic model selection
• Support Ensemble solution
• Support ENAS
• New Tuners Supports
  • Hyperband
  • Grid search

Training Services

• Support Local
• Support Remote Machine
• Support OpenPAI
• Support Kubernetes
• Support other Cloud based training services (Azure Kubernetes Service, etc.)
• Support more efficient trial job training by leveraging optimizations in system level

Examples and PR

• .ipynb samples
• Fashionmnist