updated seed

.gitignore

@@ -120,3 +120,8 @@ venv.bak/
 
 # IDEs
 .idea
+
+# seed project
+lightning_logs/
+MNIST
+.DS_Store

README.md

@@ -47,19 +47,20 @@ pip install -r requirements.txt
  Next, navigate to [Your Main Contribution (MNIST here)] and run it.   
  ```bash
 # module folder
-cd research_seed/mnist/   
+cd src/    
 
 # run module (example: mnist as your main contribution)   
-python mnist_trainer.py    
+python simplest_mnist.py    
 ```
 
 ## Main Contribution      
 List your modules here. Each module contains all code for a full system including how to run instructions.   
-- [MNIST](https://github.com/williamFalcon/pytorch-lightning-conference-seed/tree/master/research_seed/mnist)  
+- [Production MNIST](https://github.com/PyTorchLightning/pytorch-lightning-conference-seed/tree/master/src/produtcion_mnist)    
+- [Research MNIST](https://github.com/PyTorchLightning/pytorch-lightning-conference-seed/tree/master/src/research_mnist)  
 
 ## Baselines    
 List your baselines here.   
-- [MNIST_baseline](https://github.com/williamFalcon/pytorch-lightning-conference-seed/tree/master/research_seed/baselines/mnist_baseline)  
+- [Research MNIST](https://github.com/PyTorchLightning/pytorch-lightning-conference-seed/tree/master/src/research_mnist) 
 
 ### Citation   
 ```

src/baselines/README.md

-## Baselines    
-Set up a folder for each baseline. Each baseline needs a trainer and LightningModule   
-

src/baselines/mnist_baseline/README.md

-## MNIST Baseline    
-In this readme, give instructions on how to run your code.   
-
-#### CPU   
-```bash   
-python mnist_baseline_trainer.py     
-```
-
-#### Multiple-GPUs   
-```bash   
-python mnist_baseline_trainer.py --gpus '0,1,2,3'  
-```   
-
-#### On multiple nodes   
-```bash  
-python mnist_baseline_trainer.py --gpus '0,1,2,3' --nodes 4  
-```   

src/baselines/mnist_baseline/mnist_baseline.py

-"""
-This file defines the core research contribution   
-"""
-import os
-import torch
-from torch.nn import functional as F
-from torch.utils.data import DataLoader
-from torchvision.datasets import MNIST
-import torchvision.transforms as transforms
-from argparse import ArgumentParser
-
-import pytorch_lightning as pl
-
-
-class CoolSystem(pl.LightningModule):
-
-    def __init__(self, hparams):
-        super(CoolSystem, self).__init__()
-        # not the best model...
-        self.hparams = hparams
-        self.l1 = torch.nn.Linear(28 * 28, 10)
-
-    def forward(self, x):
-        return torch.relu(self.l1(x.view(x.size(0), -1)))
-
-    def training_step(self, batch, batch_idx):
-        # REQUIRED
-        x, y = batch
-        y_hat = self.forward(x)
-        return {'loss': F.cross_entropy(y_hat, y)}
-
-    def validation_step(self, batch, batch_idx):
-        # OPTIONAL
-        x, y = batch
-        y_hat = self.forward(x)
-        return {'val_loss': F.cross_entropy(y_hat, y)}
-
-    def validation_end(self, outputs):
-        # OPTIONAL
-        avg_loss = torch.stack([x['val_loss'] for x in outputs]).mean()
-        return {'avg_val_loss': avg_loss}
-
-    def configure_optimizers(self):
-        # REQUIRED
-        # can return multiple optimizers and learning_rate schedulers
-        return torch.optim.Adam(self.parameters(), lr=self.hparams.learning_rate)
-
-    def train_dataloader(self):
-        # REQUIRED
-        return DataLoader(MNIST(os.getcwd(), train=True, download=True, transform=transforms.ToTensor()), batch_size=self.hparams.batch_size)
-
-    def val_dataloader(self):
-        # OPTIONAL
-        return DataLoader(MNIST(os.getcwd(), train=True, download=True, transform=transforms.ToTensor()), batch_size=self.hparams.batch_size)
-
-    def test_dataloader(self):
-        # OPTIONAL
-        return DataLoader(MNIST(os.getcwd(), train=True, download=True, transform=transforms.ToTensor()), batch_size=self.hparams.batch_size)
-
-    @staticmethod
-    def add_model_specific_args(parent_parser):
-        """
-        Specify the hyperparams for this LightningModule
-        """
-        # MODEL specific
-        parser = ArgumentParser(parents=[parent_parser])
-        parser.add_argument('--learning_rate', default=0.02, type=float)
-        parser.add_argument('--batch_size', default=32, type=int)
-
-        # training specific (for this model)
-        parser.add_argument('--max_nb_epochs', default=2, type=int)
-
-        return parser
-

src/baselines/mnist_baseline/mnist_baseline_trainer.py

-"""
-This file runs the main training/val loop, etc... using Lightning Trainer    
-"""
-from pytorch_lightning import Trainer
-from argparse import ArgumentParser
-from research_seed.mnist.mnist import CoolSystem
-
-
-def main(hparams):
-    # init module
-    model = CoolSystem(hparams)
-
-    # most basic trainer, uses good defaults
-    trainer = Trainer(
-        max_nb_epochs=hparams.max_nb_epochs,
-        gpus=hparams.gpus,
-        nb_gpu_nodes=hparams.nodes,
-    )
-    trainer.fit(model)
-
-
-if __name__ == '__main__':
-    parser = ArgumentParser(add_help=False)
-    parser.add_argument('--gpus', type=str, default=None)
-    parser.add_argument('--nodes', type=int, default=1)
-
-    # give the module a chance to add own params
-    # good practice to define LightningModule speficic params in the module
-    parser = CoolSystem.add_model_specific_args(parser)
-
-    # parse params
-    hparams = parser.parse_args()
-
-    main(hparams)

src/research_mnist/mnist.py

@@ -46,6 +46,19 @@ class CoolSystem(pl.LightningModule):
         tensorboard_logs = {'avg_val_loss': avg_loss}
         return {'val_loss': avg_loss, 'log': tensorboard_logs}
 
+    def test_step(self, batch, batch_idx):
+        # OPTIONAL
+        x, y = batch
+        y_hat = self.forward(x)
+        return {'test_loss': F.cross_entropy(y_hat, y)}
+
+    def test_epoch_end(self, outputs):
+        # OPTIONAL
+        avg_loss = torch.stack([x['test_loss'] for x in outputs]).mean()
+
+        tensorboard_logs = {'test_val_loss': avg_loss}
+        return {'test_loss': avg_loss, 'log': tensorboard_logs}
+
     def configure_optimizers(self):
         # REQUIRED
         # can return multiple optimizers and learning_rate schedulers

src/research_mnist/mnist_trainer.py

@@ -3,7 +3,7 @@ This file runs the main training/val loop, etc... using Lightning Trainer
 """
 from pytorch_lightning import Trainer, seed_everything
 from argparse import ArgumentParser
-from srv.mnist.mnist import CoolSystem
+from src.research_mnist.mnist import CoolSystem
 
 # sets seeds for numpy, torch, etc...
 # must do for DDP to work well
@@ -17,6 +17,8 @@ def main(args):
     trainer = Trainer.from_argparse_args(args)
     trainer.fit(model)
 
+    trainer.test()
+
 
 if __name__ == '__main__':
     parser = ArgumentParser(add_help=False)

src/single_file_mnist/mnist.py → src/simplest_mnist.py

@@ -8,7 +8,6 @@ from torch.utils.data import DataLoader
 from torchvision.datasets import MNIST
 import torchvision.transforms as transforms
 from argparse import ArgumentParser
-
 import pytorch_lightning as pl
 
 pl.seed_everything(123)
@@ -18,7 +17,6 @@ class CoolSystem(pl.LightningModule):
 
     def __init__(self, hparams):
         super(CoolSystem, self).__init__()
-        # not the best model...
         self.hparams = hparams
         self.l1 = torch.nn.Linear(28 * 28, 10)
 
@@ -26,70 +24,26 @@ class CoolSystem(pl.LightningModule):
         return torch.relu(self.l1(x.view(x.size(0), -1)))
 
     def training_step(self, batch, batch_idx):
-        # REQUIRED
         x, y = batch
-        y_hat = self.forward(x)
+        y_hat = self(x)
         loss = F.cross_entropy(y_hat, y)
-
         tensorboard_logs = {'train_loss': loss}
-
         return {'loss': loss, 'log': tensorboard_logs}
 
-    def validation_step(self, batch, batch_idx):
-        # OPTIONAL
-        x, y = batch
-        y_hat = self.forward(x)
-        return {'val_loss': F.cross_entropy(y_hat, y)}
-
-    def validation_epoch_end(self, outputs):
-        # OPTIONAL
-        avg_loss = torch.stack([x['val_loss'] for x in outputs]).mean()
-
-        tensorboard_logs = {'avg_val_loss': avg_loss}
-        return {'val_loss': avg_loss, 'log': tensorboard_logs}
-
     def configure_optimizers(self):
         # REQUIRED
         # can return multiple optimizers and learning_rate schedulers
         return torch.optim.Adam(self.parameters(), lr=self.hparams.learning_rate)
 
-    def train_dataloader(self):
-        # REQUIRED
-        return DataLoader(MNIST(os.getcwd(), train=True, download=True, transform=transforms.ToTensor()), batch_size=self.hparams.batch_size)
-
-    def val_dataloader(self):
-        # OPTIONAL
-        return DataLoader(MNIST(os.getcwd(), train=True, download=True, transform=transforms.ToTensor()), batch_size=self.hparams.batch_size)
-
-    def test_dataloader(self):
-        # OPTIONAL
-        return DataLoader(MNIST(os.getcwd(), train=True, download=True, transform=transforms.ToTensor()), batch_size=self.hparams.batch_size)
-
-    @staticmethod
-    def add_model_specific_args(parent_parser):
-        """
-        Specify the hyperparams for this LightningModule
-        """
-        # MODEL specific
-        parser = ArgumentParser(parents=[parent_parser], add_help=False)
-        parser.add_argument('--learning_rate', default=0.02, type=float)
-        parser.add_argument('--batch_size', default=32, type=int)
-
-        # training specific (for this model)
-        parser.add_argument('--max_nb_epochs', default=2, type=int)
-
-        return parser
-
 
 if __name__ == '__main__':
+    train_data = DataLoader(MNIST(os.getcwd(), train=True, download=True, transform=transforms.ToTensor()), batch_size=32)
+
     parser = ArgumentParser(add_help=False)
+    parser.add_argument('--learning_rate', default=0.02, type=float)
 
     # add args from trainer
-    parser = Trainer.add_argparse_args(parser)
-
-    # give the module a chance to add own params
-    # good practice to define LightningModule speficic params in the module
-    parser = CoolSystem.add_model_specific_args(parser)
+    parser = pl.Trainer.add_argparse_args(parser)
 
     # parse params
     args = parser.parse_args()
@@ -99,6 +53,6 @@ if __name__ == '__main__':
 
     # most basic trainer, uses good defaults
     trainer = pl.Trainer.from_argparse_args(args)
-    trainer.fit(model)
+    trainer.fit(model, train_data)