An open source AutoML toolkit for automate machine learning lifecycle, including feature engineering, neural architecture search, model compression and hyper-parameter tuning.
What would you like to be added:
Are there any plans to add back the sensitivty analysis tool to v3.0 ?
Why is this needed:
It would be great to have this tool back as a debug tool, to target better the layers that are not sensible to pruning. This tool was available in the previous versions (until v2.6) but was removed later on.
Without this feature, how does current nni work:
Without this tool, it's more difficult to identify which layers can be pruned further instead of using a uniform sparsity rate for the whole network without goint into the tedious process of trial & error. Which can be a big overhead for big models with long evaluation processes.
Components that may involve changes:
the utils package
What would you like to be added: Are there any plans to add back the sensitivty analysis tool to v3.0 ?
Why is this needed: It would be great to have this tool back as a debug tool, to target better the layers that are not sensible to pruning. This tool was available in the previous versions (until v2.6) but was removed later on.
Without this feature, how does current nni work:
Without this tool, it's more difficult to identify which layers can be pruned further instead of using a uniform sparsity rate for the whole network without goint into the tedious process of trial & error. Which can be a big overhead for big models with long evaluation processes.
Components that may involve changes: the utils package
Brief description of your proposal if any: