pytorch-lightning/docs/LightningModule/RequiredTrainerInterface.md

Lightning Module interface

[Github Code]

A lightning module is a strict superclass of nn.Module, it provides a standard interface for the trainer to interact with the model.

The easiest thing to do is copy the minimal example below and modify accordingly.

Otherwise, to Define a Lightning Module, implement the following methods:

Required:

• training_step
• tng_dataloader
• configure_optimizers

Optional:

• validation_step

• validation_end

• val_dataloader

• test_dataloader

• on_save_checkpoint

• on_load_checkpoint

• update_tng_log_metrics

• add_model_specific_args

---

Minimal example

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

import pytorch_lightning as pl

class CoolModel(pl.LightningModule):

    def __init__(self):
        super(CoolModel, self).__init__()
        # not the best model...
        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_nb):
        # REQUIRED
        x, y = batch
        y_hat = self.forward(x)
        return {'loss': F.cross_entropy(y_hat, y)(y_hat, y)}

    def validation_step(self, batch, batch_nb):
        # OPTIONAL
        x, y = batch
        y_hat = self.forward(x)
        return {'val_loss': F.cross_entropy(y_hat, y)(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
        return [torch.optim.Adam(self.parameters(), lr=0.02)]

    @pl.data_loader
    def tng_dataloader(self):
        return DataLoader(MNIST(os.getcwd(), train=True, download=True, transform=transforms.ToTensor()), batch_size=32)

    @pl.data_loader
    def val_dataloader(self):
        # OPTIONAL
        # can also return a list of val dataloaders
        return DataLoader(MNIST(os.getcwd(), train=True, download=True, transform=transforms.ToTensor()), batch_size=32)

    @pl.data_loader
    def test_dataloader(self):
        # OPTIONAL
        return DataLoader(MNIST(os.getcwd(), train=True, download=True, transform=transforms.ToTensor()), batch_size=32)

---

How do these methods fit into the broader training?

The LightningModule interface is on the right. Each method corresponds to a part of a research project. Lightning automates everything not in blue.

Required Methods

training_step

def training_step(self, data_batch, batch_nb)

In this step you'd normally do the forward pass and calculate the loss for a batch. You can also do fancier things like multiple forward passes or something specific to your model.

Params

Param	description
data_batch	The output of your dataloader. A tensor, tuple or list
batch_nb	Integer displaying which batch this is

Return

Dictionary or OrderedDict

key	value	is required
loss	tensor scalar	Y
prog	Dict for progress bar display. Must have only tensors	N

Example

def training_step(self, data_batch, batch_nb):
    x, y, z = data_batch
    
    # implement your own
    out = self.forward(x)
    loss = self.loss(out, x)
    
    output = {
        'loss': loss, # required
        'prog': {'tng_loss': loss, 'batch_nb': batch_nb} # optional
    }
    
    # return a dict
    return output

---

tng_dataloader

@pl.data_loader
def tng_dataloader(self)

Called by lightning during training loop. Make sure to use the @pl.data_loader decorator, this ensures not calling this function until the data are needed.

Return

PyTorch DataLoader

Example

@pl.data_loader
def tng_dataloader(self):
    transform = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.5,), (1.0,))])
    dataset = MNIST(root='/path/to/mnist/', train=True, transform=transform, download=True)
    loader = torch.utils.data.DataLoader(
        dataset=dataset,
        batch_size=self.hparams.batch_size,
        shuffle=True
    )
    return loader

---

configure_optimizers

def configure_optimizers(self)

Set up as many optimizers and (optionally) learning rate schedulers as you need. Normally you'd need one. But in the case of GANs or something more esoteric you might have multiple. Lightning will call .backward() and .step() on each one in every epoch. If you use 16 bit precision it will also handle that.

Return

List or Tuple - List of optimizers with an optional second list of learning-rate schedulers

Example

# most cases
def configure_optimizers(self):
    opt = Adam(self.parameters(), lr=0.01)
    return [opt]
    
# gan example, with scheduler for discriminator
def configure_optimizers(self):
    generator_opt = Adam(self.model_gen.parameters(), lr=0.01)
    disriminator_opt = Adam(self.model_disc.parameters(), lr=0.02)
    discriminator_sched = CosineAnnealing(discriminator_opt, T_max=10)
    return [generator_opt, disriminator_opt], [discriminator_sched]

Optional Methods

validation_step

def validation_step(self, data_batch, batch_nb, dataloader_i)

OPTIONAL
If you don't need to validate you don't need to implement this method.

In this step you'd normally do the forward pass and calculate the loss for a batch. You can also do fancier things like multiple forward passes, calculate accuracy, or save example outputs (using self.experiment or whatever you want). Really, anything you want.

This is most likely the same as your training_step. But unlike training step, the outputs from here will go to validation_end for collation.

Params

Param	description
data_batch	The output of your dataloader. A tensor, tuple or list
batch_nb	Integer displaying which batch this is
dataloader_i	Integer displaying which dataloader this is

Return

Return	description	optional
dict	Dict of OrderedDict with metrics to display in progress bar. All keys must be tensors.	Y

Example

def validation_step(self, data_batch, batch_nb):
    x, y, z = data_batch
    
    # implement your own
    out = self.forward(x)
    loss = self.loss(out, x)
    
    # calculate acc
    labels_hat = torch.argmax(out, dim=1)
    val_acc = torch.sum(y == labels_hat).item() / (len(y) * 1.0)
    
    # all optional...
    # return whatever you need for the collation function validation_end
    output = OrderedDict({
        'val_loss': loss_val,
        'val_acc': torch.tensor(val_acc), # everything must be a tensor
    })
    
    # return an optional dict
    return output

---

validation_end

def validation_end(self, outputs)

If you didn't define a validation_step, this won't be called.

Called at the end of the validation loop with the output of each validation_step. Called once per validation dataset.

The outputs here are strictly for the progress bar. If you don't need to display anything, don't return anything.

Params

Param	description
outputs	List of outputs you defined in validation_step

Return

Return	description	optional
dict	Dict of OrderedDict with metrics to display in progress bar	Y

Example

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:
    """
    val_loss_mean = 0
    val_acc_mean = 0
    for output in outputs:
        val_loss_mean += output['val_loss']
        val_acc_mean += output['val_acc']

    val_loss_mean /= len(outputs)
    val_acc_mean /= len(outputs)
    tqdm_dic = {'val_loss': val_loss_mean.item(), 'val_acc': val_acc_mean.item()}
    return tqdm_dic

---

on_save_checkpoint

def on_save_checkpoint(self, checkpoint)

Called by lightning to checkpoint your model. Lightning saves the training state (current epoch, global_step, etc) and also saves the model state_dict. If you want to save anything else, use this method to add your own key-value pair.

Return

Nothing

Example

def on_save_checkpoint(self, checkpoint):
    # 99% of use cases you don't need to implement this method 
    checkpoint['something_cool_i_want_to_save'] = my_cool_pickable_object

---

on_load_checkpoint

def on_load_checkpoint(self, checkpoint)

Called by lightning to restore your model. Lighting auto-restores global step, epoch, etc... It also restores the model state_dict. If you saved something with on_save_checkpoint this is your chance to restore this.

Return

Nothing

Example

def on_load_checkpoint(self, checkpoint):
    # 99% of the time you don't need to implement this method
    self.something_cool_i_want_to_save = checkpoint['something_cool_i_want_to_save']

---

val_dataloader

@pl.data_loader
def tng_dataloader(self)

OPTIONAL
If you don't need a validation dataset and a validation_step, you don't need to implement this method.

Called by lightning during validation loop. Make sure to use the @pl.data_loader decorator, this ensures not calling this function until the data are needed.

Return

PyTorch DataLoader or list of PyTorch Dataloaders.

Example

@pl.data_loader
def val_dataloader(self):
    transform = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.5,), (1.0,))])
    dataset = MNIST(root='/path/to/mnist/', train=False, transform=transform, download=True)
    loader = torch.utils.data.DataLoader(
        dataset=dataset,
        batch_size=self.hparams.batch_size,
        shuffle=True
    )
    
    return loader

# can also return multiple dataloaders   
@pl.data_loader
def val_dataloader(self):
    return [loader_a, loader_b, ..., loader_n]   

---

test_dataloader

@pl.data_loader
def test_dataloader(self)

OPTIONAL
If you don't need a test dataset and a test_step, you don't need to implement this method.

Called by lightning during test loop. Make sure to use the @pl.data_loader decorator, this ensures not calling this function until the data are needed.

Return

PyTorch DataLoader

Example

@pl.data_loader
def test_dataloader(self):
    transform = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.5,), (1.0,))])
    dataset = MNIST(root='/path/to/mnist/', train=False, transform=transform, download=True)
    loader = torch.utils.data.DataLoader(
        dataset=dataset,
        batch_size=self.hparams.batch_size,
        shuffle=True
    )
    
    return loader

---

update_tng_log_metrics

def update_tng_log_metrics(self, logs)

Called by lightning right before it logs metrics for this batch. This is a chance to ammend or add to the metrics about to be logged.

Return

Dict

Example

def update_tng_log_metrics(self, logs):
    # modify or add to logs
    return logs

---

add_model_specific_args

@staticmethod
def add_model_specific_args(parent_parser, root_dir)

Lightning has a list of default argparse commands. This method is your chance to add or modify commands specific to your model. The hyperparameter argument parser is available anywhere in your model by calling self.hparams.

Return

An argument parser

Example

@staticmethod
def add_model_specific_args(parent_parser, root_dir):
    parser = HyperOptArgumentParser(strategy=parent_parser.strategy, parents=[parent_parser])

    # param overwrites
    # parser.set_defaults(gradient_clip=5.0)

    # network params
    parser.opt_list('--drop_prob', default=0.2, options=[0.2, 0.5], type=float, tunable=False)
    parser.add_argument('--in_features', default=28*28)
    parser.add_argument('--out_features', default=10)
    parser.add_argument('--hidden_dim', default=50000) # use 500 for CPU, 50000 for GPU to see speed difference

    # data
    parser.add_argument('--data_root', default=os.path.join(root_dir, 'mnist'), type=str)

    # training params (opt)
    parser.opt_list('--learning_rate', default=0.001, type=float, options=[0.0001, 0.0005, 0.001, 0.005],
                    tunable=False)
    parser.opt_list('--batch_size', default=256, type=int, options=[32, 64, 128, 256], tunable=False)
    parser.opt_list('--optimizer_name', default='adam', type=str, options=['adam'], tunable=False)
    return parser