""" Example template for defining a system """ import os from collections import OrderedDict import torch.nn as nn from torchvision.datasets import MNIST import torchvision.transforms as transforms import torch import torch.nn.functional as F from test_tube import HyperOptArgumentParser from torch import optim from torch.utils.data import DataLoader from torch.utils.data.distributed import DistributedSampler import pytorch_lightning as pl from pytorch_lightning.root_module.root_module import LightningModule class LightningTemplateModel(LightningModule): """ Sample model to show how to define a template """ def __init__(self, hparams): """ Pass in parsed HyperOptArgumentParser to the model :param hparams: """ # init superclass super(LightningTemplateModel, self).__init__() self.hparams = hparams self.batch_size = hparams.batch_size # if you specify an example input, the summary will show input/output for each layer self.example_input_array = torch.rand(5, 28 * 28) # build model self.__build_model() # --------------------- # MODEL SETUP # --------------------- def __build_model(self): """ Layout model :return: """ self.c_d1 = nn.Linear(in_features=self.hparams.in_features, out_features=self.hparams.hidden_dim) self.c_d1_bn = nn.BatchNorm1d(self.hparams.hidden_dim) self.c_d1_drop = nn.Dropout(self.hparams.drop_prob) self.c_d2 = nn.Linear(in_features=self.hparams.hidden_dim, out_features=self.hparams.out_features) # --------------------- # TRAINING # --------------------- def forward(self, x): """ No special modification required for lightning, define as you normally would :param x: :return: """ x = self.c_d1(x) x = torch.tanh(x) x = self.c_d1_bn(x) x = self.c_d1_drop(x) x = self.c_d2(x) logits = F.log_softmax(x, dim=1) return logits def loss(self, labels, logits): nll = F.nll_loss(logits, labels) return nll def training_step(self, batch, batch_idx): """ Lightning calls this inside the training loop :param batch: :return: """ # forward pass x, y = batch x = x.view(x.size(0), -1) y_hat = self.forward(x) # calculate loss loss_val = self.loss(y, y_hat) # in DP mode (default) make sure if result is scalar, there's another dim in the beginning if self.trainer.use_dp: loss_val = loss_val.unsqueeze(0) output = OrderedDict({ 'loss': loss_val }) # can also return just a scalar instead of a dict (return loss_val) return output def validation_step(self, batch, batch_idx): """ Lightning calls this inside the validation loop :param batch: :return: """ x, y = batch x = x.view(x.size(0), -1) y_hat = self.forward(x) loss_val = self.loss(y, y_hat) # acc labels_hat = torch.argmax(y_hat, dim=1) val_acc = torch.sum(y == labels_hat).item() / (len(y) * 1.0) val_acc = torch.tensor(val_acc) if self.on_gpu: val_acc = val_acc.cuda(loss_val.device.index) # in DP mode (default) make sure if result is scalar, there's another dim in the beginning if self.trainer.use_dp: loss_val = loss_val.unsqueeze(0) val_acc = val_acc.unsqueeze(0) output = OrderedDict({ 'val_loss': loss_val, 'val_acc': val_acc, }) # can also return just a scalar instead of a dict (return loss_val) return output def validation_end(self, outputs): """ Called at the end of validation to aggregate outputs :param outputs: list of individual outputs of each validation step :return: """ # if returned a scalar from validation_step, outputs is a list of tensor scalars # we return just the average in this case (if we want) # return torch.stack(outputs).mean() val_loss_mean = 0 val_acc_mean = 0 for output in outputs: val_loss = output['val_loss'] # reduce manually when using dp if self.trainer.use_dp: val_loss = torch.mean(val_loss) val_loss_mean += val_loss # reduce manually when using dp val_acc = output['val_acc'] if self.trainer.use_dp or self.trainer.use_ddp2: val_acc = torch.mean(val_acc) val_acc_mean += val_acc val_loss_mean /= len(outputs) val_acc_mean /= len(outputs) tqdm_dict = {'val_loss': val_loss_mean, 'val_acc': val_acc_mean} result = {'progress_bar': tqdm_dict} return result # --------------------- # TRAINING SETUP # --------------------- def configure_optimizers(self): """ return whatever optimizers we want here :return: list of optimizers """ optimizer = optim.Adam(self.parameters(), lr=self.hparams.learning_rate) scheduler = optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=10) return [optimizer], [scheduler] def __dataloader(self, train): # init data generators transform = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.5,), (1.0,))]) dataset = MNIST(root=self.hparams.data_root, train=train, transform=transform, download=True) # when using multi-node (ddp) we need to add the datasampler train_sampler = None batch_size = self.hparams.batch_size if self.use_ddp: train_sampler = DistributedSampler(dataset, rank=self.trainer.proc_rank) batch_size = batch_size // self.trainer.world_size # scale batch size should_shuffle = train_sampler is None loader = DataLoader( dataset=dataset, batch_size=batch_size, shuffle=should_shuffle, sampler=train_sampler ) return loader @pl.data_loader def train_dataloader(self): print('training data loader called') return self.__dataloader(train=True) @pl.data_loader def val_dataloader(self): print('val data loader called') return self.__dataloader(train=False) @pl.data_loader def test_dataloader(self): print('test data loader called') return self.__dataloader(train=False) @staticmethod def add_model_specific_args(parent_parser, root_dir): # pragma: no cover """ Parameters you define here will be available to your model through self.hparams :param parent_parser: :param root_dir: :return: """ parser = HyperOptArgumentParser(strategy=parent_parser.strategy, parents=[parent_parser]) # param overwrites # parser.set_defaults(gradient_clip_val=5.0) # network params parser.add_argument('--in_features', default=28 * 28, type=int) parser.add_argument('--out_features', default=10, type=int) # use 500 for CPU, 50000 for GPU to see speed difference parser.add_argument('--hidden_dim', default=50000, type=int) parser.opt_list('--drop_prob', default=0.2, options=[0.2, 0.5], type=float, tunable=True) parser.opt_list('--learning_rate', default=0.001 * 8, type=float, options=[0.0001, 0.0005, 0.001], tunable=True) # data parser.add_argument('--data_root', default=os.path.join(root_dir, 'mnist'), type=str) # training params (opt) parser.opt_list('--optimizer_name', default='adam', type=str, options=['adam'], tunable=False) # if using 2 nodes with 4 gpus each the batch size here # (256) will be 256 / (2*8) = 16 per gpu parser.opt_list('--batch_size', default=256 * 8, type=int, options=[32, 64, 128, 256], tunable=False, help='batch size will be divided over all gpus being used across all nodes') return parser