One-shot Experiments on Retiarii¶
Before reading this tutorial, we highly recommend you to first go through the tutorial of how to define a model space.
Model Search with One-shot Trainer¶
With a defined model space, users can explore the space in two ways. One is using strategy and single-arch evaluator as demonstrated here. The other is using one-shot trainer, which consumes much less computational resource compared to the first one. In this tutorial we focus on this one-shot approach. The principle of one-shot approach is combining all the models in a model space into one big model (usually called super-model or super-graph). It takes charge of both search, training and testing, by training and evaluating this big model.
We list the supported one-shot trainers here:
DARTS trainer
ENAS trainer
ProxylessNAS trainer
Single-path (random) trainer
See API reference for detailed usages. Here, we show an example to use DARTS trainer manually.
from nni.retiarii.oneshot.pytorch import DartsTrainer
trainer = DartsTrainer(
model=model,
loss=criterion,
metrics=lambda output, target: accuracy(output, target, topk=(1,)),
optimizer=optim,
num_epochs=args.epochs,
dataset=dataset_train,
batch_size=args.batch_size,
log_frequency=args.log_frequency,
unrolled=args.unrolled
)
trainer.fit()
final_architecture = trainer.export()
Format of the exported architecture. TBD.
One-shot experiment can be visualized with NAS UI, please refer to here for the usage guidance. Note that NAS visualization is under intensive development.
Customize a New One-shot Trainer¶
One-shot trainers should inherit nni.retiarii.oneshot.BaseOneShotTrainer
, and need to implement fit()
(used to conduct the fitting and searching process) and export()
method (used to return the searched best architecture).
Writing a one-shot trainer is very different to single-arch evaluator. First of all, there are no more restrictions on init method arguments, any Python arguments are acceptable. Secondly, the model fed into one-shot trainers might be a model with Retiarii-specific modules, such as LayerChoice and InputChoice. Such model cannot directly forward-propagate and trainers need to decide how to handle those modules.
A typical example is DartsTrainer, where learnable-parameters are used to combine multiple choices in LayerChoice. Retiarii provides ease-to-use utility functions for module-replace purposes, namely replace_layer_choice
, replace_input_choice
. A simplified example is as follows:
from nni.retiarii.oneshot import BaseOneShotTrainer
from nni.retiarii.oneshot.pytorch import replace_layer_choice, replace_input_choice
class DartsLayerChoice(nn.Module):
def __init__(self, layer_choice):
super(DartsLayerChoice, self).__init__()
self.name = layer_choice.key
self.op_choices = nn.ModuleDict(layer_choice.named_children())
self.alpha = nn.Parameter(torch.randn(len(self.op_choices)) * 1e-3)
def forward(self, *args, **kwargs):
op_results = torch.stack([op(*args, **kwargs) for op in self.op_choices.values()])
alpha_shape = [-1] + [1] * (len(op_results.size()) - 1)
return torch.sum(op_results * F.softmax(self.alpha, -1).view(*alpha_shape), 0)
class DartsTrainer(BaseOneShotTrainer):
def __init__(self, model, loss, metrics, optimizer):
self.model = model
self.loss = loss
self.metrics = metrics
self.num_epochs = 10
self.nas_modules = []
replace_layer_choice(self.model, DartsLayerChoice, self.nas_modules)
... # init dataloaders and optimizers
def fit(self):
for i in range(self.num_epochs):
for (trn_X, trn_y), (val_X, val_y) in zip(self.train_loader, self.valid_loader):
self.train_architecture(val_X, val_y)
self.train_model_weight(trn_X, trn_y)
@torch.no_grad()
def export(self):
result = dict()
for name, module in self.nas_modules:
if name not in result:
result[name] = select_best_of_module(module)
return result
The full code of DartsTrainer is available to Retiarii source code. Please have a check at DartsTrainer.