[Retiarii] Rewrite trainer with PyTorch Lightning

dependencies/recommended.txt

@@ -2,13 +2,11 @@
 
 -f https://download.pytorch.org/whl/torch_stable.html
 tensorflow
-
-# PyTorch 1.7 has compatibility issue with model compression.
-# Check for MacOS because this file is used on all platforms.
 torch == 1.6.0+cpu ; sys_platform != "darwin"
 torch == 1.6.0 ; sys_platform == "darwin"
 torchvision == 0.7.0+cpu ; sys_platform != "darwin"
 torchvision == 0.7.0 ; sys_platform == "darwin"
+pytorch-lightning
 onnx
 peewee
 graphviz

dependencies/recommended_legacy.txt

@@ -2,6 +2,11 @@
 tensorflow == 1.15.4
 torch == 1.5.1+cpu
 torchvision == 0.6.1+cpu
+
+# It will install pytorch-lightning 0.8.x and unit tests won't work.
+# Latest version has conflict with tensorboard and tensorflow 1.x.
+pytorch-lightning
+
 keras == 2.1.6
 onnx
 peewee

docs/en_US/NAS/retiarii/ApiReference.rst

@@ -42,10 +42,16 @@ Graph Mutation APIs
 Trainers
 --------
 
-..  autoclass:: nni.retiarii.trainer.pytorch.PyTorchImageClassificationTrainer
+..  autoclass:: nni.retiarii.trainer.FunctionalTrainer
     :members:
 
-..  autoclass:: nni.retiarii.trainer.pytorch.PyTorchMultiModelTrainer
+..  autoclass:: nni.retiarii.trainer.pytorch.lightning.LightningModule
+    :members:
+
+..  autoclass:: nni.retiarii.trainer.pytorch.lightning.Classification
+    :members:
+
+..  autoclass:: nni.retiarii.trainer.pytorch.lightning.Regression
     :members:
 
 Oneshot Trainers
@@ -75,8 +81,8 @@ Strategies
 Retiarii Experiments
 --------------------
 
-..  autoclass:: nni.retiarii.experiment.RetiariiExperiment
+..  autoclass:: nni.retiarii.experiment.pytorch.RetiariiExperiment
     :members:
 
-..  autoclass:: nni.retiarii.experiment.RetiariiExeConfig
+..  autoclass:: nni.retiarii.experiment.pytorch.RetiariiExeConfig
     :members:

docs/en_US/NAS/retiarii/Tutorial.rst

@@ -149,7 +149,7 @@ Create a Trainer and Exploration Strategy
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
 **Classic search approach:**
-In this approach, trainer is for training each explored model, while strategy is for sampling the models. Both trainer and strategy are required to explore the model space.
+In this approach, trainer is for training each explored model, while strategy is for sampling the models. Both trainer and strategy are required to explore the model space. We recommend PyTorch-Lightning to write the full training process.
 
 **Oneshot (weight-sharing) search approach:**
 In this approach, users only need a oneshot trainer, because this trainer takes charge of both search and training.
@@ -163,10 +163,10 @@ In the following table, we listed the available trainers and strategies.
   * - Trainer
     - Strategy
     - Oneshot Trainer
-  * - PyTorchImageClassificationTrainer
+  * - Classification
     - TPEStrategy
     - DartsTrainer
-  * - PyTorchMultiModelTrainer
+  * - Regression
     - RandomStrategy
     - EnasTrainer
   * - 
@@ -182,15 +182,20 @@ Here is a simple example of using trainer and strategy.
 
 .. code-block:: python
 
-  trainer = PyTorchImageClassificationTrainer(base_model,   
-    dataset_cls="MNIST",
-    dataset_kwargs={"root": "data/mnist", "download": True},
-    dataloader_kwargs={"batch_size": 32},
-    optimizer_kwargs={"lr": 1e-3},
-    trainer_kwargs={"max_epochs": 1})
-  simple_startegy = RandomStrategy()
+  import nni.retiarii.trainer.pytorch.lightning as pl
+  from nni.retiarii import blackbox
+  from torchvision import transforms
 
-Users can refer to `this document <./WriteTrainer.rst>`__ for how to write a new trainer, and refer to `this document <./WriteStrategy.rst>`__ for how to write a new strategy.
+  transform = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.1307,), (0.3081,))])
+  train_dataset = blackbox(MNIST, root='data/mnist', train=True, download=True, transform=transform)
+  test_dataset = blackbox(MNIST, root='data/mnist', train=False, download=True, transform=transform)
+  lightning = pl.Classification(train_dataloader=pl.DataLoader(train_dataset, batch_size=100),
+                                val_dataloaders=pl.DataLoader(test_dataset, batch_size=100),
+                                max_epochs=10)
+
+.. Note:: For NNI to capture the dataset and dataloader and distribute it across different runs, please wrap your dataset with ``blackbox`` and use ``pl.DataLoader`` instead of ``torch.utils.data.DataLoader``. See ``blackbox_module`` section below for details.
+
+Users can refer to `API reference <./ApiReference.rst>`__ on detailed usage of trainer. "`write a trainer <./WriteTrainer.rst>`__" for how to write a new trainer, and refer to `this document <./WriteStrategy.rst>`__ for how to write a new strategy.
 
 Set up an Experiment
 ^^^^^^^^^^^^^^^^^^^^

docs/en_US/NAS/retiarii/WriteTrainer.rst

@@ -3,59 +3,114 @@ Customize A New Trainer
 
 Trainers are necessary to evaluate the performance of new explored models. In NAS scenario, this further divides into two use cases:
 
-1. **Classic trainers**: trainers that are used to train and evaluate one single model.
+1. **Single-arch trainers**: trainers that are used to train and evaluate one single model.
 2. **One-shot trainers**: trainers that handle training and searching simultaneously, from an end-to-end perspective.
 
-Classic trainers
-----------------
+Single-arch trainers
+--------------------
 
-All classic trainers need to inherit ``nni.retiarii.trainer.BaseTrainer``, implement the ``fit`` method and decorated with ``@register_trainer`` if it is intended to be used together with Retiarii. The decorator serialize the trainer that is used and its argument to fit for the requirements of NNI.
+With PyTorch-Lightning
+^^^^^^^^^^^^^^^^^^^^^^
 
-The init function of trainer should take model as its first argument, and the rest of the arguments should be named (``*args`` and ``**kwargs`` may not work as expected) and JSON serializable. This means, currently, passing a complex object like ``torchvision.datasets.ImageNet()`` is not supported. Trainer should use NNI standard API to communicate with tuning algorithms. This includes ``nni.report_intermediate_result`` for periodical metrics and ``nni.report_final_result`` for final metrics.
+It's recommended to write training code in PyTorch-Lightning style, that is, to write a LightningModule that defines all elements needed for training (e.g., loss function, optimizer) and to define a trainer that takes (optional) dataloaders to execute the training. Before that, please read the `document of PyTorch-lightning <https://pytorch-lightning.readthedocs.io/>` to learn the basic concepts and components provided by PyTorch-lightning.
+
+In pratice, writing a new training module in NNI should inherit ``nni.retiarii.trainer.pytorch.lightning.LightningModule``, which has a ``set_model`` that will be called after ``__init__`` to save the candidate model (generated by strategy) as ``self.model``. The rest of the process (like ``training_step``) should be the same as writing any other lightning module. Trainers should also communicate with strategies via two API calls (``nni.report_intermediate_result`` for periodical metrics and ``nni.report_final_result`` for final metrics), added in ``on_validation_epoch_end`` and ``teardown`` respectively. 
 
 An example is as follows:
 
 .. code-block::python
 
-    from nni.retiarii import register_trainer
-    from nni.retiarii.trainer import BaseTrainer
+    from nni.retiarii.trainer.pytorch.lightning import LightningModule  # please import this one
 
-    @register_trainer
-    class MnistTrainer(BaseTrainer):
-        def __init__(self, model, optimizer_class_name='SGD', learning_rate=0.1):
+    @blackbox_module
+    class AutoEncoder(LightningModule):
+        def __init__(self):
             super().__init__()
-            self.model = model
-            self.criterion = nn.CrossEntropyLoss()
-            self.train_dataset = MNIST(train=True)
-            self.valid_dataset = MNIST(train=False)
-            self.optimizer = getattr(torch.optim, optimizer_class_name)(lr=learning_rate)
-
-        def validate():
-            pass
-
-        def fit(self) -> None:
-            for i in range(10):  # number of epochs:
-                for x, y in DataLoader(self.dataset):
-                    self.optimizer.zero_grad()
-                    pred = self.model(x)
-                    loss = self.criterion(pred, y)
-                    loss.backward()
-                    self.optimizer.step()
-            acc = self.validate()  # get validation accuracy
-            nni.report_final_result(acc)
+            self.decoder = nn.Sequential(
+                nn.Linear(3, 64),
+                nn.ReLU(),
+                nn.Linear(64, 28*28)
+            )
+
+        def forward(self, x):
+            embedding = self.model(x)  # let's search for encoder
+            return embedding
+
+        def training_step(self, batch, batch_idx):
+            # training_step defined the train loop.
+            # It is independent of forward
+            x, y = batch
+            x = x.view(x.size(0), -1)
+            z = self.model(x)  # model is the one that is searched for
+            x_hat = self.decoder(z)
+            loss = F.mse_loss(x_hat, x)
+            # Logging to TensorBoard by default
+            self.log('train_loss', loss)
+            return loss
+
+        def validation_step(self, batch, batch_idx):
+            x, y = batch
+            x = x.view(x.size(0), -1)
+            z = self.model(x)
+            x_hat = self.decoder(z)
+            loss = F.mse_loss(x_hat, x)
+            self.log('val_loss', loss)
+
+        def configure_optimizers(self):
+            optimizer = torch.optim.Adam(self.parameters(), lr=1e-3)
+            return optimizer
+
+        def on_validation_epoch_end(self):
+            nni.report_intermediate_result(self.trainer.callback_metrics['val_loss'].item())
+
+        def teardown(self, stage):
+            if stage == 'fit':
+                nni.report_final_result(self.trainer.callback_metrics['val_loss'].item())
+
+Then, users need to wrap everything (including LightningModule, trainer and dataloaders) into a ``Lightning`` object, and pass this object into a Retiarii experiment.
+
+.. code-block::python
+
+    import nni.retiarii.trainer.pytorch.lightning as pl
+    from nni.retiarii.experiment.pytorch import RetiariiExperiment
+
+    lightning = pl.Lightning(AutoEncoder(),
+                             pl.Trainer(max_epochs=10),
+                             train_dataloader=pl.DataLoader(train_dataset, batch_size=100),
+                             val_dataloaders=pl.DataLoader(test_dataset, batch_size=100))
+    experiment = RetiariiExperiment(base_model, lightning, mutators, strategy)
+
+With FunctionalTrainer
+^^^^^^^^^^^^^^^^^^^^^^
+
+There is another way to customize a new trainer with functional APIs, which provides more flexibility. Users only need to write a fit function that wraps everything. This function takes one positional arguments (model) and possible keyword arguments. In this way, users get everything under their control, but exposes less information to the framework and thus fewer opportunities for possible optimization. An example is as belows:
+
+.. code-block::python
+
+    from nni.retiarii.trainer import FunctionalTrainer
+    from nni.retiarii.experiment.pytorch import RetiariiExperiment
+
+    def fit(model, dataloader):
+        train(model, dataloader)
+        acc = test(model, dataloader)
+        nni.report_final_result(acc)
+
+    trainer = FunctionalTrainer(fit, dataloader=DataLoader(foo, bar))
+    experiment = RetiariiExperiment(base_model, trainer, mutators, strategy)
+
 
 One-shot trainers
 -----------------
 
-One-shot trainers should inheirt ``nni.retiarii.trainer.BaseOneShotTrainer``, which is basically same as ``BaseTrainer``, but only with one extra method ``export()``, which is expected to return the searched best architecture.
+One-shot trainers should inheirt ``nni.retiarii.trainer.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 classic trainers. First of all, there are no more restrictions on init method arguments, any Python arguments are acceptable. Secondly, the model feeded 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: 
 
 .. code-block::python
 
-    from nni.retiarii.trainer import BaseOneShotTrainer
+    from nni.retiarii.trainer.pytorch import BaseOneShotTrainer
     from nni.retiarii.trainer.pytorch.utils import replace_layer_choice, replace_input_choice
 
 

nni/retiarii/__init__.py

@@ -2,4 +2,4 @@
 from .graph import *
 from .execution import *
 from .mutator import *
-from .utils import blackbox, blackbox_module, register_trainer
+from .utils import blackbox, blackbox_module, json_dump, json_dumps, json_load, json_loads, register_trainer

nni/retiarii/execution/base.py

@@ -2,31 +2,29 @@
 import os
 import random
 import string
-from typing import Dict, Any, List
+from typing import Dict, List
 
 from .interface import AbstractExecutionEngine, AbstractGraphListener
 from .. import codegen, utils
-from ..graph import Model, ModelStatus, MetricData
+from ..graph import Model, ModelStatus, MetricData, TrainingConfig
 from ..integration_api import send_trial, receive_trial_parameters, get_advisor
 
 _logger = logging.getLogger(__name__)
 
 class BaseGraphData:
-    def __init__(self, model_script: str, training_module: str, training_kwargs: Dict[str, Any]) -> None:
+    def __init__(self, model_script: str, training_config: TrainingConfig) -> None:
         self.model_script = model_script
-        self.training_module = training_module
-        self.training_kwargs = training_kwargs
+        self.training_config = training_config
 
     def dump(self) -> dict:
         return {
             'model_script': self.model_script,
-            'training_module': self.training_module,
-            'training_kwargs': self.training_kwargs
+            'training_config': self.training_config
         }
 
     @staticmethod
-    def load(data):
-        return BaseGraphData(data['model_script'], data['training_module'], data['training_kwargs'])
+    def load(data) -> 'BaseGraphData':
+        return BaseGraphData(data['model_script'], data['training_config'])
 
 
 class BaseExecutionEngine(AbstractExecutionEngine):
@@ -57,8 +55,7 @@ def __init__(self) -> None:
 
     def submit_models(self, *models: Model) -> None:
         for model in models:
-            data = BaseGraphData(codegen.model_to_pytorch_script(model),
-                                 model.training_config.module, model.training_config.kwargs)
+            data = BaseGraphData(codegen.model_to_pytorch_script(model), model.training_config)
             self._running_models[send_trial(data.dump())] = model
 
     def register_graph_listener(self, listener: AbstractGraphListener) -> None:
@@ -105,11 +102,10 @@ def trial_execute_graph(cls) -> None:
         """
         graph_data = BaseGraphData.load(receive_trial_parameters())
         random_str = ''.join(random.choice(string.ascii_uppercase + string.digits) for _ in range(6))
-        file_name = f'_generated_model_{random_str}.py'
+        file_name = f'_generated_model/{random_str}.py'
+        os.makedirs(os.path.dirname(file_name), exist_ok=True)
         with open(file_name, 'w') as f:
             f.write(graph_data.model_script)
-        trainer_cls = utils.import_(graph_data.training_module)
-        model_cls = utils.import_(f'_generated_model_{random_str}._model')
-        trainer_instance = trainer_cls(model=model_cls(), **graph_data.training_kwargs)
-        trainer_instance.fit()
-        os.remove(file_name)
+        model_cls = utils.import_(f'_generated_model.{random_str}._model')
+        graph_data.training_config._execute(model_cls)
+        os.remove(file_name)

nni/retiarii/execution/cgo_engine.py

@@ -44,7 +44,7 @@ def submit_models(self, *models: List[Model]) -> None:
         phy_models_and_placements = self._assemble(logical)
         for model, placement, grouped_models in phy_models_and_placements:
             data = BaseGraphData(codegen.model_to_pytorch_script(model, placement=placement),
-                                 model.training_config.module, model.training_config.kwargs)
+                                 model.training_config)
             for m in grouped_models:
                 self._original_models[m.model_id] = m
                 self._original_model_to_multi_model[m.model_id] = model