gents.model.vae package

Module contents

class gents.model.vae.KoVAE(seq_len: int, seq_dim: int, condition: str | None = None, latent_dim: int = 16, hidden_size: int = 20, num_layers: int = 3, batch_norm: bool = True, w_rec: float = 1.0, w_kl: float = 0.007, w_pred_prior: float = 0.005, pinv_solver: bool = False, koopman_nstep: int = 1, irregular_ts: bool = False, lr: float = 0.0007, weight_decay: float = 0.0, **kwargs)

Bases: BaseModel

KoVAE: Koopman Variational Autoencoder for both Regular and Irregular Time Series

Adapted from the official codes

Note

KoVAE allows for irregular data.

Parameters:

• seq_len (int) – Target sequence length

• seq_dim (int) – Target sequence dimension, for univariate time series, set as 1

• condition (str, optional) – Given conditions, allowing [None]. Defaults to None.

• latent_dim (int, optional) – Latent dimension for z. Defaults to 16.

• hidden_size (int, optional) – Hidden size. Defaults to 20.

• num_layers (int, optional) – RNN layers. Defaults to 3.

• batch_norm (bool, optional) – Batch norm or not. Defaults to True.

• w_rec (float, optional) – Loss weight of reconstruction. Defaults to 1.0.

• w_kl (float, optional) – Loss weight of KL divergance. Defaults to 0.007.

• w_pred_prior (float, optional) – Loss weight of prior distribution learning. Defaults to 0.005.

• pinv_solver (bool, optional) – Directly calculate pseudoinverse or not. Defaults to False.

• koopman_nstep (int, optional) – N step ahead dynamic learning. Defaults to 1.

• irregular_ts (bool, optional) – Whether the input data is irregular. Defaults to False.

• lr (float, optional) – Learning rate. Defaults to 7e-4.

• weight_decay (float, optional) – Weight decay. Defaults to 0.0.

• **kwargs – Arbitrary keyword arguments, e.g. obs_len, class_num, etc.

ALLOW_CONDITION = [None]

configure_optimizers()

Choose what optimizers and learning-rate schedulers to use in your optimization. Normally you’d need one. But in the case of GANs or similar you might have multiple. Optimization with multiple optimizers only works in the manual optimization mode.

Returns:

Any of these 6 options.

• Single optimizer.

• List or Tuple of optimizers.

• Two lists - The first list has multiple optimizers, and the second has multiple LR schedulers (or multiple lr_scheduler_config).

• Dictionary, with an "optimizer" key, and (optionally) a "lr_scheduler" key whose value is a single LR scheduler or lr_scheduler_config.

• None - Fit will run without any optimizer.

The lr_scheduler_config is a dictionary which contains the scheduler and its associated configuration. The default configuration is shown below.

lr_scheduler_config = {
    # REQUIRED: The scheduler instance
    "scheduler": lr_scheduler,
    # The unit of the scheduler's step size, could also be 'step'.
    # 'epoch' updates the scheduler on epoch end whereas 'step'
    # updates it after a optimizer update.
    "interval": "epoch",
    # How many epochs/steps should pass between calls to
    # `scheduler.step()`. 1 corresponds to updating the learning
    # rate after every epoch/step.
    "frequency": 1,
    # Metric to to monitor for schedulers like `ReduceLROnPlateau`
    "monitor": "val_loss",
    # If set to `True`, will enforce that the value specified 'monitor'
    # is available when the scheduler is updated, thus stopping
    # training if not found. If set to `False`, it will only produce a warning
    "strict": True,
    # If using the `LearningRateMonitor` callback to monitor the
    # learning rate progress, this keyword can be used to specify
    # a custom logged name
    "name": None,
}

When there are schedulers in which the .step() method is conditioned on a value, such as the torch.optim.lr_scheduler.ReduceLROnPlateau scheduler, Lightning requires that the lr_scheduler_config contains the keyword "monitor" set to the metric name that the scheduler should be conditioned on.

# The ReduceLROnPlateau scheduler requires a monitor
def configure_optimizers(self):
    optimizer = Adam(...)
    return {
        "optimizer": optimizer,
        "lr_scheduler": {
            "scheduler": ReduceLROnPlateau(optimizer, ...),
            "monitor": "metric_to_track",
            "frequency": "indicates how often the metric is updated",
            # If "monitor" references validation metrics, then "frequency" should be set to a
            # multiple of "trainer.check_val_every_n_epoch".
        },
    }


# In the case of two optimizers, only one using the ReduceLROnPlateau scheduler
def configure_optimizers(self):
    optimizer1 = Adam(...)
    optimizer2 = SGD(...)
    scheduler1 = ReduceLROnPlateau(optimizer1, ...)
    scheduler2 = LambdaLR(optimizer2, ...)
    return (
        {
            "optimizer": optimizer1,
            "lr_scheduler": {
                "scheduler": scheduler1,
                "monitor": "metric_to_track",
            },
        },
        {"optimizer": optimizer2, "lr_scheduler": scheduler2},
    )

Metrics can be made available to monitor by simply logging it using self.log('metric_to_track', metric_val) in your LightningModule.

Note

Some things to know:

• Lightning calls .backward() and .step() automatically in case of automatic optimization.

• If a learning rate scheduler is specified in configure_optimizers() with key "interval" (default “epoch”) in the scheduler configuration, Lightning will call the scheduler’s .step() method automatically in case of automatic optimization.

• If you use 16-bit precision (precision=16), Lightning will automatically handle the optimizer.

• If you use torch.optim.LBFGS, Lightning handles the closure function automatically for you.

• If you use multiple optimizers, you will have to switch to ‘manual optimization’ mode and step them yourself.

• If you need to control how often the optimizer steps, override the optimizer_step() hook.

forward(x, time=None, final_index=None)

Same as torch.nn.Module.forward().

Parameters:

• *args – Whatever you decide to pass into the forward method.

• **kwargs – Keyword arguments are also possible.

Returns:

Your model’s output

training_step(batch, batch_idx)

Here you compute and return the training loss and some additional metrics for e.g. the progress bar or logger.

Parameters:

• batch – The output of your data iterable, normally a DataLoader.

• batch_idx – The index of this batch.

• dataloader_idx – The index of the dataloader that produced this batch. (only if multiple dataloaders used)

Returns:

• Tensor - The loss tensor

• dict - A dictionary which can include any keys, but must include the key 'loss' in the case of automatic optimization.

• None - In automatic optimization, this will skip to the next batch (but is not supported for multi-GPU, TPU, or DeepSpeed). For manual optimization, this has no special meaning, as returning the loss is not required.

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 model specific.

Example:

def training_step(self, batch, batch_idx):
    x, y, z = batch
    out = self.encoder(x)
    loss = self.loss(out, x)
    return loss

To use multiple optimizers, you can switch to ‘manual optimization’ and control their stepping:

def __init__(self):
    super().__init__()
    self.automatic_optimization = False


# Multiple optimizers (e.g.: GANs)
def training_step(self, batch, batch_idx):
    opt1, opt2 = self.optimizers()

    # do training_step with encoder
    ...
    opt1.step()
    # do training_step with decoder
    ...
    opt2.step()

Note

When accumulate_grad_batches > 1, the loss returned here will be automatically normalized by accumulate_grad_batches internally.

validation_step(batch, batch_idx)

Operates on a single batch of data from the validation set. In this step you’d might generate examples or calculate anything of interest like accuracy.

Parameters:

• batch – The output of your data iterable, normally a DataLoader.

• batch_idx – The index of this batch.

• dataloader_idx – The index of the dataloader that produced this batch. (only if multiple dataloaders used)

Returns:

• Tensor - The loss tensor

• dict - A dictionary. Can include any keys, but must include the key 'loss'.

• None - Skip to the next batch.

# if you have one val dataloader:
def validation_step(self, batch, batch_idx): ...


# if you have multiple val dataloaders:
def validation_step(self, batch, batch_idx, dataloader_idx=0): ...

Examples:

# CASE 1: A single validation dataset
def validation_step(self, batch, batch_idx):
    x, y = batch

    # implement your own
    out = self(x)
    loss = self.loss(out, y)

    # log 6 example images
    # or generated text... or whatever
    sample_imgs = x[:6]
    grid = torchvision.utils.make_grid(sample_imgs)
    self.logger.experiment.add_image('example_images', grid, 0)

    # calculate acc
    labels_hat = torch.argmax(out, dim=1)
    val_acc = torch.sum(y == labels_hat).item() / (len(y) * 1.0)

    # log the outputs!
    self.log_dict({'val_loss': loss, 'val_acc': val_acc})

If you pass in multiple val dataloaders, validation_step() will have an additional argument. We recommend setting the default value of 0 so that you can quickly switch between single and multiple dataloaders.

# CASE 2: multiple validation dataloaders
def validation_step(self, batch, batch_idx, dataloader_idx=0):
    # dataloader_idx tells you which dataset this is.
    ...

Note

If you don’t need to validate you don’t need to implement this method.

Note

When the validation_step() is called, the model has been put in eval mode and PyTorch gradients have been disabled. At the end of validation, the model goes back to training mode and gradients are enabled.

class gents.model.vae.TimeVAE(seq_len: int, seq_dim: int, condition: str | None = None, latent_dim: int = 128, hidden_size_list=[64, 128, 256], w_kl: float = 0.005, trend_poly: int = 2, custom_seas: List[Tuple[int, int]] | None = None, use_residual_conn: bool = True, lr: float = 0.001, weight_decay: float = 1e-05, **kwargs)

Bases: VanillaVAE

TimeVAE for time series generation.

Adapted from the official codes

Note

The orignial codes are based on Tensorflow, we adapt the source codes into pytorch.

Parameters:

• seq_len (int) – Target sequence length

• seq_dim (int) – Target sequence dimension, for univariate time series, set as 1

• condition (str, optional) – Given condition type, should be one of ALLOW_CONDITION. Defaults to None.

• latent_dim (int, optional) – Latent variable dimension. Defaults to 128.

• hidden_size_list (list, optional) – Hidden size for Conv encoder and decoder. Defaults to [64, 128, 256].

• w_kl (float, optional) – Loss weight of KL div. Defaults to 1e-4.

• trend_poly (int, optional) – integer for number of orders for trend component. e.g. setting trend_poly = 2 will include linear and quadratic term.

• custom_seas (List[Tuple[int, int]], optional) – list of tuples of (num_seasons, len_per_season). num_seasons: number of seasons per cycle. len_per_season: number of epochs (time-steps) per season.

• use_residual_conn (bool, optional) – boolean value indicating whether to use a residual connection for reconstruction in addition to trend, generic and custom seasonalities.

• lr (float, optional) – Learning rate. Defaults to 1e-3.

• weight_decay (float, optional) – Weight decay. Defaults to 1e-5.

• **kwargs – Arbitrary keyword arguments, e.g. obs_len, class_num, etc.

ALLOW_CONDITION = [None]

class gents.model.vae.TimeVQVAE(seq_len: int, seq_dim: int, condition: str | None = None, resnet_init_dim: int = 4, hidden_size: int = 128, n_fft: int = 4, n_resnet_blocks: int = 2, downsampled_width_l: int = 8, downsampled_width_h: int = 32, codebook_size: int = 1024, lr: float = 0.001, stage_split: float = 0.5, cfg_scale: float = 0.5, prior_model_l_config: Dict[str, Any] = {'emb_dropout': 0.3, 'ff_mult': 1, 'heads': 2, 'hidden_dim': 128, 'model_dropout': 0.3, 'n_layers': 4, 'p_unconditional': 0.2, 'use_rmsnorm': True}, prior_model_h_config: Dict[str, Any] = {'emb_dropout': 0.3, 'ff_mult': 1, 'heads': 1, 'hidden_dim': 32, 'model_dropout': 0.3, 'n_layers': 1, 'p_unconditional': 0.2, 'use_rmsnorm': True}, choice_temperatures: Dict[str, Any] = {'hf': 0, 'lf': 10}, dec_iter_step: Dict[str, int] = {'hf': 10, 'lf': 10}, **kwargs)

Bases: BaseModel

TimeVQVAE for time series generation.

Adapted from the official codes

Parameters:

• seq_len (int) – Target sequence length

• seq_dim (int) – Target sequence dimension, for univariate time series, set as 1

• condition (str, optional) – Given conditions, should be one of ALLOW_CONDITION. Defaults to None.

• resnet_init_dim (int, optional) – Initial d_model of resnet. Defaults to 4.

• hidden_size (int, optional) – Hidden size. Defaults to 128.

• n_fft (int, optional) – Size of Fourier transform. Defaults to 4.

• n_resnet_blocks (int, optional) – Blocks of Resnet. Defaults to 2.

• downsampled_width_l (int, optional) – Low-frequency downsampled width. Defaults to 8.

• downsampled_width_h (int, optional) – High-frequency downsampled width. Defaults to 32.

• codebook_size (int, optional) – VQ codebook. Defaults to 1024.

• lr (float, optional) – Learning rate. Defaults to 1e-3.

• stage_split (float, optional) – Training stage splite ratio, [0, 1]. Defaults to 0.5.

• cfg_scale (float, optional) – Classifier free guidance rate for conditional generation, [0, 1]. Defaults to 0.5.

• prior_model_l_config (Dict[str, Any], optional) – Prior model config for low-frequency. Defaults to dict( hidden_dim=128, n_layers=4, heads=2, ff_mult=1, use_rmsnorm=True, p_unconditional=0.2, model_dropout=0.3, emb_dropout=0.3, ).

• prior_model_h_config (Dict[str, Any], optional) – Prior model config for high-frequency. Defaults to dict( hidden_dim=32, n_layers=1, heads=1, ff_mult=1, use_rmsnorm=True, p_unconditional=0.2, model_dropout=0.3, emb_dropout=0.3, ).

• choice_temperatures (Dict[str, Any], optional) – Temperatures for randomness for low-freq and high-freq. Defaults to {“lf”: 10, “hf”: 0}.

• dec_iter_step (Dict[str, int], optional) – Decoding iteration steps for low-freq and high-freq. Defaults to {“lf”: 10, “hf”: 10}.

• **kwargs – Arbitrary keyword arguments, e.g. obs_len, class_num, etc.

ALLOW_CONDITION = [None, 'class']

configure_optimizers()

Choose what optimizers and learning-rate schedulers to use in your optimization. Normally you’d need one. But in the case of GANs or similar you might have multiple. Optimization with multiple optimizers only works in the manual optimization mode.

Returns:

Any of these 6 options.

• Single optimizer.

• List or Tuple of optimizers.

• Two lists - The first list has multiple optimizers, and the second has multiple LR schedulers (or multiple lr_scheduler_config).

• Dictionary, with an "optimizer" key, and (optionally) a "lr_scheduler" key whose value is a single LR scheduler or lr_scheduler_config.

• None - Fit will run without any optimizer.

The lr_scheduler_config is a dictionary which contains the scheduler and its associated configuration. The default configuration is shown below.

lr_scheduler_config = {
    # REQUIRED: The scheduler instance
    "scheduler": lr_scheduler,
    # The unit of the scheduler's step size, could also be 'step'.
    # 'epoch' updates the scheduler on epoch end whereas 'step'
    # updates it after a optimizer update.
    "interval": "epoch",
    # How many epochs/steps should pass between calls to
    # `scheduler.step()`. 1 corresponds to updating the learning
    # rate after every epoch/step.
    "frequency": 1,
    # Metric to to monitor for schedulers like `ReduceLROnPlateau`
    "monitor": "val_loss",
    # If set to `True`, will enforce that the value specified 'monitor'
    # is available when the scheduler is updated, thus stopping
    # training if not found. If set to `False`, it will only produce a warning
    "strict": True,
    # If using the `LearningRateMonitor` callback to monitor the
    # learning rate progress, this keyword can be used to specify
    # a custom logged name
    "name": None,
}

When there are schedulers in which the .step() method is conditioned on a value, such as the torch.optim.lr_scheduler.ReduceLROnPlateau scheduler, Lightning requires that the lr_scheduler_config contains the keyword "monitor" set to the metric name that the scheduler should be conditioned on.

# The ReduceLROnPlateau scheduler requires a monitor
def configure_optimizers(self):
    optimizer = Adam(...)
    return {
        "optimizer": optimizer,
        "lr_scheduler": {
            "scheduler": ReduceLROnPlateau(optimizer, ...),
            "monitor": "metric_to_track",
            "frequency": "indicates how often the metric is updated",
            # If "monitor" references validation metrics, then "frequency" should be set to a
            # multiple of "trainer.check_val_every_n_epoch".
        },
    }


# In the case of two optimizers, only one using the ReduceLROnPlateau scheduler
def configure_optimizers(self):
    optimizer1 = Adam(...)
    optimizer2 = SGD(...)
    scheduler1 = ReduceLROnPlateau(optimizer1, ...)
    scheduler2 = LambdaLR(optimizer2, ...)
    return (
        {
            "optimizer": optimizer1,
            "lr_scheduler": {
                "scheduler": scheduler1,
                "monitor": "metric_to_track",
            },
        },
        {"optimizer": optimizer2, "lr_scheduler": scheduler2},
    )

Metrics can be made available to monitor by simply logging it using self.log('metric_to_track', metric_val) in your LightningModule.

Note

Some things to know:

• Lightning calls .backward() and .step() automatically in case of automatic optimization.

• If a learning rate scheduler is specified in configure_optimizers() with key "interval" (default “epoch”) in the scheduler configuration, Lightning will call the scheduler’s .step() method automatically in case of automatic optimization.

• If you use 16-bit precision (precision=16), Lightning will automatically handle the optimizer.

• If you use torch.optim.LBFGS, Lightning handles the closure function automatically for you.

• If you use multiple optimizers, you will have to switch to ‘manual optimization’ mode and step them yourself.

• If you need to control how often the optimizer steps, override the optimizer_step() hook.

training_step(batch, batch_idx)

Two-stage training for TimeVQVAE.

Stage 1: Learning Vector Quantization

Stage 2: Prior Learning

Note

Note that TimeVQVAE is limited by max_steps instead of max_epochs.

validation_step(batch, batch_idx)

Operates on a single batch of data from the validation set. In this step you’d might generate examples or calculate anything of interest like accuracy.

Parameters:

• batch – The output of your data iterable, normally a DataLoader.

• batch_idx – The index of this batch.

• dataloader_idx – The index of the dataloader that produced this batch. (only if multiple dataloaders used)

Returns:

• Tensor - The loss tensor

• dict - A dictionary. Can include any keys, but must include the key 'loss'.

• None - Skip to the next batch.

# if you have one val dataloader:
def validation_step(self, batch, batch_idx): ...


# if you have multiple val dataloaders:
def validation_step(self, batch, batch_idx, dataloader_idx=0): ...

Examples:

# CASE 1: A single validation dataset
def validation_step(self, batch, batch_idx):
    x, y = batch

    # implement your own
    out = self(x)
    loss = self.loss(out, y)

    # log 6 example images
    # or generated text... or whatever
    sample_imgs = x[:6]
    grid = torchvision.utils.make_grid(sample_imgs)
    self.logger.experiment.add_image('example_images', grid, 0)

    # calculate acc
    labels_hat = torch.argmax(out, dim=1)
    val_acc = torch.sum(y == labels_hat).item() / (len(y) * 1.0)

    # log the outputs!
    self.log_dict({'val_loss': loss, 'val_acc': val_acc})

If you pass in multiple val dataloaders, validation_step() will have an additional argument. We recommend setting the default value of 0 so that you can quickly switch between single and multiple dataloaders.

# CASE 2: multiple validation dataloaders
def validation_step(self, batch, batch_idx, dataloader_idx=0):
    # dataloader_idx tells you which dataset this is.
    ...

Note

If you don’t need to validate you don’t need to implement this method.

Note

When the validation_step() is called, the model has been put in eval mode and PyTorch gradients have been disabled. At the end of validation, the model goes back to training mode and gradients are enabled.

class gents.model.vae.VanillaVAE(seq_len: int, seq_dim: int, condition: str | None = None, latent_dim: int = 128, backbone: str = 'mlp', backbone_params: dict | None = None, w_kl: float = 0.0001, lr: float = 0.001, weight_decay: float = 1e-05, **kwargs)

Bases: BaseModel

Vanilla Variational Autoencoder (VAE) model with flexible backbone.

For conditional generation, an extra encoder is used for embedding conditions.

Parameters:

• seq_len (int) – Target sequence length

• seq_dim (int) – Target sequence dimension, for univariate time series, set as 1

• condition (str, optional) – Given condition type, should be one of ALLOW_CONDITION. Defaults to None.

• latent_dim (int, optional) – Latent variable dimension. Defaults to 128.

• backbone (str, optional) – Backbone type, one of ‘mlp’, ‘rnn’, ‘transformer’. Defaults to ‘mlp’.

• backbone_params (dict, optional) – Backbone-specific hyperparameters. Defaults to None. For ‘mlp’: hidden_size_list (list) For ‘rnn’: hidden_size (int), num_layers (int), dropout (float), rnn_type (str) For ‘transformer’: d_model (int), nhead (int), num_layers (int), dim_feedforward (int), dropout (float)

• w_kl (float, optional) – Loss weight of KL div. Defaults to 1e-4.

• lr (float, optional) – Learning rate. Defaults to 1e-3.

• weight_decay (float, optional) – Weight decay. Defaults to 1e-5.

• **kwargs – Arbitrary keyword arguments, e.g. obs_len, class_num, etc.

ALLOW_CONDITION = [None, 'predict', 'impute', 'class', 'super_resolution']

configure_optimizers()

Choose what optimizers and learning-rate schedulers to use in your optimization. Normally you’d need one. But in the case of GANs or similar you might have multiple. Optimization with multiple optimizers only works in the manual optimization mode.

Returns:

Any of these 6 options.

• Single optimizer.

• List or Tuple of optimizers.

• Two lists - The first list has multiple optimizers, and the second has multiple LR schedulers (or multiple lr_scheduler_config).

• Dictionary, with an "optimizer" key, and (optionally) a "lr_scheduler" key whose value is a single LR scheduler or lr_scheduler_config.

• None - Fit will run without any optimizer.

The lr_scheduler_config is a dictionary which contains the scheduler and its associated configuration. The default configuration is shown below.

lr_scheduler_config = {
    # REQUIRED: The scheduler instance
    "scheduler": lr_scheduler,
    # The unit of the scheduler's step size, could also be 'step'.
    # 'epoch' updates the scheduler on epoch end whereas 'step'
    # updates it after a optimizer update.
    "interval": "epoch",
    # How many epochs/steps should pass between calls to
    # `scheduler.step()`. 1 corresponds to updating the learning
    # rate after every epoch/step.
    "frequency": 1,
    # Metric to to monitor for schedulers like `ReduceLROnPlateau`
    "monitor": "val_loss",
    # If set to `True`, will enforce that the value specified 'monitor'
    # is available when the scheduler is updated, thus stopping
    # training if not found. If set to `False`, it will only produce a warning
    "strict": True,
    # If using the `LearningRateMonitor` callback to monitor the
    # learning rate progress, this keyword can be used to specify
    # a custom logged name
    "name": None,
}

When there are schedulers in which the .step() method is conditioned on a value, such as the torch.optim.lr_scheduler.ReduceLROnPlateau scheduler, Lightning requires that the lr_scheduler_config contains the keyword "monitor" set to the metric name that the scheduler should be conditioned on.

# The ReduceLROnPlateau scheduler requires a monitor
def configure_optimizers(self):
    optimizer = Adam(...)
    return {
        "optimizer": optimizer,
        "lr_scheduler": {
            "scheduler": ReduceLROnPlateau(optimizer, ...),
            "monitor": "metric_to_track",
            "frequency": "indicates how often the metric is updated",
            # If "monitor" references validation metrics, then "frequency" should be set to a
            # multiple of "trainer.check_val_every_n_epoch".
        },
    }


# In the case of two optimizers, only one using the ReduceLROnPlateau scheduler
def configure_optimizers(self):
    optimizer1 = Adam(...)
    optimizer2 = SGD(...)
    scheduler1 = ReduceLROnPlateau(optimizer1, ...)
    scheduler2 = LambdaLR(optimizer2, ...)
    return (
        {
            "optimizer": optimizer1,
            "lr_scheduler": {
                "scheduler": scheduler1,
                "monitor": "metric_to_track",
            },
        },
        {"optimizer": optimizer2, "lr_scheduler": scheduler2},
    )

Metrics can be made available to monitor by simply logging it using self.log('metric_to_track', metric_val) in your LightningModule.

Note

Some things to know:

• Lightning calls .backward() and .step() automatically in case of automatic optimization.

• If a learning rate scheduler is specified in configure_optimizers() with key "interval" (default “epoch”) in the scheduler configuration, Lightning will call the scheduler’s .step() method automatically in case of automatic optimization.

• If you use 16-bit precision (precision=16), Lightning will automatically handle the optimizer.

• If you use torch.optim.LBFGS, Lightning handles the closure function automatically for you.

• If you use multiple optimizers, you will have to switch to ‘manual optimization’ mode and step them yourself.

• If you need to control how often the optimizer steps, override the optimizer_step() hook.

training_step(batch, batch_idx)

Here you compute and return the training loss and some additional metrics for e.g. the progress bar or logger.

Parameters:

• batch – The output of your data iterable, normally a DataLoader.

• batch_idx – The index of this batch.

• dataloader_idx – The index of the dataloader that produced this batch. (only if multiple dataloaders used)

Returns:

• Tensor - The loss tensor

• dict - A dictionary which can include any keys, but must include the key 'loss' in the case of automatic optimization.

• None - In automatic optimization, this will skip to the next batch (but is not supported for multi-GPU, TPU, or DeepSpeed). For manual optimization, this has no special meaning, as returning the loss is not required.

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 model specific.

Example:

def training_step(self, batch, batch_idx):
    x, y, z = batch
    out = self.encoder(x)
    loss = self.loss(out, x)
    return loss

To use multiple optimizers, you can switch to ‘manual optimization’ and control their stepping:

def __init__(self):
    super().__init__()
    self.automatic_optimization = False


# Multiple optimizers (e.g.: GANs)
def training_step(self, batch, batch_idx):
    opt1, opt2 = self.optimizers()

    # do training_step with encoder
    ...
    opt1.step()
    # do training_step with decoder
    ...
    opt2.step()

Note

When accumulate_grad_batches > 1, the loss returned here will be automatically normalized by accumulate_grad_batches internally.

validation_step(batch, batch_idx)

Operates on a single batch of data from the validation set. In this step you’d might generate examples or calculate anything of interest like accuracy.

Parameters:

• batch – The output of your data iterable, normally a DataLoader.

• batch_idx – The index of this batch.

• dataloader_idx – The index of the dataloader that produced this batch. (only if multiple dataloaders used)

Returns:

• Tensor - The loss tensor

• dict - A dictionary. Can include any keys, but must include the key 'loss'.

• None - Skip to the next batch.

# if you have one val dataloader:
def validation_step(self, batch, batch_idx): ...


# if you have multiple val dataloaders:
def validation_step(self, batch, batch_idx, dataloader_idx=0): ...

Examples:

# CASE 1: A single validation dataset
def validation_step(self, batch, batch_idx):
    x, y = batch

    # implement your own
    out = self(x)
    loss = self.loss(out, y)

    # log 6 example images
    # or generated text... or whatever
    sample_imgs = x[:6]
    grid = torchvision.utils.make_grid(sample_imgs)
    self.logger.experiment.add_image('example_images', grid, 0)

    # calculate acc
    labels_hat = torch.argmax(out, dim=1)
    val_acc = torch.sum(y == labels_hat).item() / (len(y) * 1.0)

    # log the outputs!
    self.log_dict({'val_loss': loss, 'val_acc': val_acc})

If you pass in multiple val dataloaders, validation_step() will have an additional argument. We recommend setting the default value of 0 so that you can quickly switch between single and multiple dataloaders.

# CASE 2: multiple validation dataloaders
def validation_step(self, batch, batch_idx, dataloader_idx=0):
    # dataloader_idx tells you which dataset this is.
    ...

Note

If you don’t need to validate you don’t need to implement this method.

Note

When the validation_step() is called, the model has been put in eval mode and PyTorch gradients have been disabled. At the end of validation, the model goes back to training mode and gradients are enabled.