matluster

@liuzhe-lz Any comments on this? P.S. When I looked at [pyright](https://github.com/microsoft/pyright) the other day, I noticed an interesting way, that is to publish the ts package on npm, and maintain...

> I'm trying to perform exactly https://github.com/microsoft/nni/tree/master/examples/nas/oneshot/proxylessnas example. Using the same search space on mobilenet-like model and imagenet dataset. Is it possible to do the same exercise on the strategy.Proxyless()?...

@mashu You shouldn't pickle the large TSV data in any case. You should put the logic of reading data into a function, and pickle that function instead; or use `nni.trace`...

It might look ugly, but my point here is that `train_dataset` and `test_dataset` shouldn't be directly pickled. Instead, their constructing process should be pickled. You might be shooting for something...

CategoricalMultiple is not supported for TPE strategy. But I think we can do better by handling the special case where `n_chosen=1` for CategoricalMultiple in `as_legacy_dict()`. The `as_legacy_dict()` was meant to...

There is a GumbelDARTS, which has a same implementation as the legacy FBNet.

Lower learning rate might be helpful.

It appears that the blob access is set to private for some reason. Please try again as I just turned on public access.

Thanks @jungin500. I think Microsoft has some security regulations that is preventing us from making the blob public. They might have periodically-run script that closes the public access if the...

1. You need a new (probably custom defined profiler). 2. NNI NAS 3.0 is an upgraded version of Retiarii NAS. The old version is out-of-date and won't be maintained any...