ENAS and DRATS search space zoo #2589
Conversation
| from nni.nas.pytorch import mutables | ||
|
|
||
|
|
||
| class DartsSearchSpace(nn.Module): |
There was a problem hiding this comment.
I think DARTS search space is a cell rather than a full model. Suggest to changing into DartsStackedCells that accepts a cell class to build a full model.
There was a problem hiding this comment.
I think DARTS search space is a cell rather than a full model. Suggest to changing into
DartsStackedCellsthat accepts a cell class to build a full model.
please review new commit 99c841b
| return out | ||
|
|
||
|
|
||
| class StdConv(nn.Module): |
There was a problem hiding this comment.
Suggest to keep all convolutions in a separate file.
There was a problem hiding this comment.
Custom ops will be implemented later. I dont think it is appropriate to expose these builtin ops now.
| """ | ||
| def __init__(self, C_in, C_out, kernel_size, stride, padding, affine=True): | ||
| super().__init__() | ||
| self.net = nn.Sequential( |
There was a problem hiding this comment.
You can actually inheirt nn.Sequential here and use self.add_module.
| the number of cells contained in this network | ||
| factory_func: function | ||
| return an instance for certain cell structure | ||
| user should pass in `__init__` of demand cell class |
There was a problem hiding this comment.
Should write
``__init__``
Suggest to make the signature clearer here. For example, it can be a factory function with the following parameters, and return an object callable with the following parameters...
There was a problem hiding this comment.
I listed some required parameters. But I believe the best way is referring to an example.
| --- | ||
| n_nodes: int | ||
| the number of nodes contained in this cell | ||
| channels_pp: int channels_pp: int |
tabVersion
changed the title
initial commit for search space zoo