Fix minor issues in DARTS and add a simple CIFAR10 example (with random mutator) #1776
Conversation
…-refactor-sync-1120 # Conflicts: # examples/nas/.gitignore # examples/nas/darts/model.py # examples/nas/darts/retrain.py # examples/nas/enas/macro.py # src/sdk/pynni/nni/nas/pytorch/base_mutator.py # src/sdk/pynni/nni/nas/pytorch/darts/trainer.py # src/sdk/pynni/nni/nas/pytorch/enas/trainer.py # src/sdk/pynni/nni/nas/pytorch/fixed.py # src/sdk/pynni/nni/nas/pytorch/mutables.py # src/sdk/pynni/nni/nas/pytorch/pdarts/mutator.py # src/sdk/pynni/nni/nas/pytorch/pdarts/trainer.py # src/sdk/pynni/nni/nas/pytorch/trainer.py
… dev-nas-refactor-sync-1120
… into dev-nas-refactor-sync-1120
| @@ -0,0 +1,25 @@ | |||
| import torch | |||
There was a problem hiding this comment.
License will be included in a PR after cutting branch, as discussed before.
| @@ -0,0 +1,25 @@ | |||
| import torch | |||
examples/nas/random/train.py
Outdated
|
|
||
| if __name__ == "__main__": | ||
| transform = transforms.Compose( | ||
| [transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]) |
There was a problem hiding this comment.
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))?
You could reference here:
https://github.com/kuangliu/pytorch-cifar/blob/master/main.py
There was a problem hiding this comment.
This is an example to demonstrate the power of our interface, not for obtaining the high accuracy. The code is copied from here. I think putting a notebook is better but that will be in notebooks folder?
examples/nas/random/train.py
Outdated
| print('[%d, %5d] loss: %.3f' % (epoch + 1, i + 1, running_loss / 100)) | ||
| running_loss = 0.0 | ||
|
|
||
| print("Finished training. Final architecture:") |
There was a problem hiding this comment.
could you provide some examples about how to use this architecture to retrain?
There was a problem hiding this comment.
Example for retrain is already in darts. This is a simple example.
Will fix merging issues tomorrow.