lfxai.models.images module

class AutoEncoderMnist(encoder: EncoderMnist, decoder: DecoderMnist, latent_dim: int, input_pert: callable, name: str = 'model', loss_f: callable = MSELoss())

Bases: Module

_backward_hooks: Dict[int, Callable]

_buffers: Dict[str, Optional[Tensor]]

_forward_hooks: Dict[int, Callable]

_forward_pre_hooks: Dict[int, Callable]

_is_full_backward_hook: Optional[bool]

_load_state_dict_post_hooks: Dict[int, Callable]

_load_state_dict_pre_hooks: Dict[int, Callable]

_modules: Dict[str, Optional[Module]]

_non_persistent_buffers_set: Set[str]

_parameters: Dict[str, Optional[Parameter]]

_state_dict_hooks: Dict[int, Callable]

fit(device: device, train_loader: DataLoader, test_loader: DataLoader, save_dir: Path, n_epoch: int = 30, patience: int = 10, checkpoint_interval: int = -1) → None

forward(x)

Forward pass of model.

Parameters:

xtorch.Tensor

Batch of data. Shape (batch_size, n_chan, height, width)

load_metadata(directory: Path) → dict

Load the metadata of a training directory.

Parameters:

directorypathlib.Path

Path to folder where model is saved. For example ‘./experiments/mnist’.

save(directory: Path) → None

Save a model and corresponding metadata.

Parameters:

directorypathlib.Path

Path to the directory where to save the data.

save_metadata(directory: Path, **kwargs) → None

Load the metadata of a training directory.

Parameters:

directory: string

Path to folder where to save model. For example ‘./experiments/mnist’.

kwargs:

Additional arguments to json.dump

test_epoch(device: device, dataloader: DataLoader)

train_epoch(device: device, dataloader: DataLoader, optimizer: Optimizer) → ndarray

training: bool

class BetaTcVaeMnist(latent_dims: int = 10, beta: int = 1)

Bases: Module

_backward_hooks: Dict[int, Callable]

_buffers: Dict[str, Optional[Tensor]]

_forward_hooks: Dict[int, Callable]

_forward_pre_hooks: Dict[int, Callable]

_is_full_backward_hook: Optional[bool]

_load_state_dict_post_hooks: Dict[int, Callable]

_load_state_dict_pre_hooks: Dict[int, Callable]

_modules: Dict[str, Optional[Module]]

_non_persistent_buffers_set: Set[str]

_parameters: Dict[str, Optional[Parameter]]

_state_dict_hooks: Dict[int, Callable]

fit(device: device, train_loader: DataLoader, test_loader: DataLoader, n_epoch: int = 30) → None

forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

latent_sample(mu, logvar)

loss(recon_x: Tensor, x: Tensor, mu: Tensor, logvar: Tensor, z: Tensor, dataset_size: int) → Tensor

test_epoch(device: device, dataloader: DataLoader)

train_epoch(device: device, dataloader: DataLoader, optimizer: Optimizer) → ndarray

training: bool

class BetaVaeMnist(latent_dims: int = 10, beta: int = 1)

Bases: Module

_backward_hooks: Dict[int, Callable]

_buffers: Dict[str, Optional[Tensor]]

_forward_hooks: Dict[int, Callable]

_forward_pre_hooks: Dict[int, Callable]

_is_full_backward_hook: Optional[bool]

_load_state_dict_post_hooks: Dict[int, Callable]

_load_state_dict_pre_hooks: Dict[int, Callable]

_modules: Dict[str, Optional[Module]]

_non_persistent_buffers_set: Set[str]

_parameters: Dict[str, Optional[Parameter]]

_state_dict_hooks: Dict[int, Callable]

fit(device: device, train_loader: DataLoader, test_loader: DataLoader, n_epoch: int = 30) → None

forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

latent_sample(mu, logvar)

loss(recon_x: Tensor, x: Tensor, mu: Tensor, logvar: Tensor, dataset_size: int) → Tensor

test_epoch(device: device, dataloader: DataLoader) → ndarray

train_epoch(device: device, dataloader: DataLoader, optimizer: Optimizer) → ndarray

training: bool

class ClassifierLatent(latent_dims: int)

Bases: Module

_backward_hooks: Dict[int, Callable]

_buffers: Dict[str, Optional[Tensor]]

_forward_hooks: Dict[int, Callable]

_forward_pre_hooks: Dict[int, Callable]

_is_full_backward_hook: Optional[bool]

_load_state_dict_post_hooks: Dict[int, Callable]

_load_state_dict_pre_hooks: Dict[int, Callable]

_modules: Dict[str, Optional[Module]]

_non_persistent_buffers_set: Set[str]

_parameters: Dict[str, Optional[Parameter]]

_state_dict_hooks: Dict[int, Callable]

forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class ClassifierMnist(encoder: EncoderMnist, latent_dim: int, name: str)

Bases: Module

_backward_hooks: Dict[int, Callable]

_buffers: Dict[str, Optional[Tensor]]

_forward_hooks: Dict[int, Callable]

_forward_pre_hooks: Dict[int, Callable]

_is_full_backward_hook: Optional[bool]

_load_state_dict_post_hooks: Dict[int, Callable]

_load_state_dict_pre_hooks: Dict[int, Callable]

_modules: Dict[str, Optional[Module]]

_non_persistent_buffers_set: Set[str]

_parameters: Dict[str, Optional[Parameter]]

_state_dict_hooks: Dict[int, Callable]

fit(device: device, train_loader: DataLoader, test_loader: DataLoader, save_dir: Path, n_epoch: int = 30, patience: int = 10) → None

forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

load_metadata(directory: Path) → dict

Load the metadata of a training directory.

Parameters:

directorypathlib.Path

Path to folder where model is saved. For example ‘./experiments/mnist’.

save(directory: Path) → None

Save a model and corresponding metadata.

Parameters:

directorypathlib.Path

Path to the directory where to save the data.

save_metadata(directory: Path, **kwargs) → None

Load the metadata of a training directory.

Parameters:

directory: string

Path to folder where to save model. For example ‘./experiments/mnist’.

kwargs:

Additional arguments to json.dump

test_epoch(device: device, dataloader: DataLoader)

train_epoch(device: device, dataloader: DataLoader, optimizer: Optimizer) → ndarray

training: bool

class DecoderBurgess(img_size, latent_dim=10)

Bases: Module

_backward_hooks: Dict[int, Callable]

_buffers: Dict[str, Optional[Tensor]]

_forward_hooks: Dict[int, Callable]

_forward_pre_hooks: Dict[int, Callable]

_is_full_backward_hook: Optional[bool]

_load_state_dict_post_hooks: Dict[int, Callable]

_load_state_dict_pre_hooks: Dict[int, Callable]

_modules: Dict[str, Optional[Module]]

_non_persistent_buffers_set: Set[str]

_parameters: Dict[str, Optional[Parameter]]

_state_dict_hooks: Dict[int, Callable]

forward(z)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class DecoderMnist(encoded_space_dim)

Bases: Module

_backward_hooks: Dict[int, Callable]

_buffers: Dict[str, Optional[Tensor]]

_forward_hooks: Dict[int, Callable]

_forward_pre_hooks: Dict[int, Callable]

_is_full_backward_hook: Optional[bool]

_load_state_dict_post_hooks: Dict[int, Callable]

_load_state_dict_pre_hooks: Dict[int, Callable]

_modules: Dict[str, Optional[Module]]

_non_persistent_buffers_set: Set[str]

_parameters: Dict[str, Optional[Parameter]]

_state_dict_hooks: Dict[int, Callable]

forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class EncoderBurgess(img_size, latent_dim=10)

Bases: Module

_backward_hooks: Dict[int, Callable]

_buffers: Dict[str, Optional[Tensor]]

_forward_hooks: Dict[int, Callable]

_forward_pre_hooks: Dict[int, Callable]

_is_full_backward_hook: Optional[bool]

_load_state_dict_post_hooks: Dict[int, Callable]

_load_state_dict_pre_hooks: Dict[int, Callable]

_modules: Dict[str, Optional[Module]]

_non_persistent_buffers_set: Set[str]

_parameters: Dict[str, Optional[Parameter]]

_state_dict_hooks: Dict[int, Callable]

forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

mu(x)

training: bool

class EncoderMnist(encoded_space_dim)

Bases: Module

_backward_hooks: Dict[int, Callable]

_buffers: Dict[str, Optional[Tensor]]

_forward_hooks: Dict[int, Callable]

_forward_pre_hooks: Dict[int, Callable]

_is_full_backward_hook: Optional[bool]

_load_state_dict_post_hooks: Dict[int, Callable]

_load_state_dict_pre_hooks: Dict[int, Callable]

_modules: Dict[str, Optional[Module]]

_non_persistent_buffers_set: Set[str]

_parameters: Dict[str, Optional[Parameter]]

_state_dict_hooks: Dict[int, Callable]

forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class SimCLR(base_encoder, projection_dim=128)

Bases: Module

_backward_hooks: Dict[int, Callable]

_buffers: Dict[str, Optional[Tensor]]

_forward_hooks: Dict[int, Callable]

_forward_pre_hooks: Dict[int, Callable]

_is_full_backward_hook: Optional[bool]

_load_state_dict_post_hooks: Dict[int, Callable]

_load_state_dict_pre_hooks: Dict[int, Callable]

_modules: Dict[str, Optional[Module]]

_non_persistent_buffers_set: Set[str]

_parameters: Dict[str, Optional[Parameter]]

_state_dict_hooks: Dict[int, Callable]

fit(args: DictConfig, device: device) → None

forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

static get_color_distortion(s=0.5)

static get_lr(step, total_steps, lr_max, lr_min)

Compute learning rate according to cosine annealing schedule.

static nt_xent(x, t=0.5)

training: bool

class VAE(img_size: tuple, encoder: EncoderBurgess, decoder: DecoderBurgess, latent_dim: int, loss_f: BaseVAELoss, name: str = 'model')

Bases: Module

_backward_hooks: Dict[int, Callable]

_buffers: Dict[str, Optional[Tensor]]

_forward_hooks: Dict[int, Callable]

_forward_pre_hooks: Dict[int, Callable]

_is_full_backward_hook: Optional[bool]

_load_state_dict_post_hooks: Dict[int, Callable]

_load_state_dict_pre_hooks: Dict[int, Callable]

_modules: Dict[str, Optional[Module]]

_non_persistent_buffers_set: Set[str]

_parameters: Dict[str, Optional[Parameter]]

_state_dict_hooks: Dict[int, Callable]

fit(device: device, train_loader: DataLoader, test_loader: DataLoader, save_dir: Path, n_epoch: int = 30, patience: int = 10) → None

forward(x)

Forward pass of model.

Parameters:

xtorch.Tensor

Batch of data. Shape (batch_size, n_chan, height, width)

load_metadata(directory: Path) → dict

Load the metadata of a training directory.

Parameters:

directorypathlib.Path

Path to folder where model is saved. For example ‘./experiments/mnist’.

reparameterize(mean, logvar)

Samples from a normal distribution using the reparameterization trick.

Parameters:

meantorch.Tensor

Mean of the normal distribution. Shape (batch_size, latent_dim)

logvartorch.Tensor

Diagonal log variance of the normal distribution. Shape (batch_size, latent_dim)

sample_latent(x)

Returns a sample from the latent distribution.

Parameters:

xtorch.Tensor

Batch of data. Shape (batch_size, n_chan, height, width)

save(directory: Path) → None

Save a model and corresponding metadata.

Parameters:

directorypathlib.Path

Path to the directory where to save the data.

save_metadata(directory: Path, **kwargs) → None

Load the metadata of a training directory.

Parameters:

directory: string

Path to folder where to save model. For example ‘./experiments/mnist’.

kwargs:

Additional arguments to json.dump

test_epoch(device: device, dataloader: DataLoader)

train_epoch(device: device, dataloader: DataLoader, optimizer: Optimizer) → ndarray

training: bool

class VarDecoderMnist(c: int = 64, latent_dims: int = 10)

Bases: Module

_backward_hooks: Dict[int, Callable]

_buffers: Dict[str, Optional[Tensor]]

_forward_hooks: Dict[int, Callable]

_forward_pre_hooks: Dict[int, Callable]

_is_full_backward_hook: Optional[bool]

_load_state_dict_post_hooks: Dict[int, Callable]

_load_state_dict_pre_hooks: Dict[int, Callable]

_modules: Dict[str, Optional[Module]]

_non_persistent_buffers_set: Set[str]

_parameters: Dict[str, Optional[Parameter]]

_state_dict_hooks: Dict[int, Callable]

forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class VarEncoderMnist(c: int = 64, latent_dims: int = 10)

Bases: Module

_backward_hooks: Dict[int, Callable]

_buffers: Dict[str, Optional[Tensor]]

_forward_hooks: Dict[int, Callable]

_forward_pre_hooks: Dict[int, Callable]

_is_full_backward_hook: Optional[bool]

_load_state_dict_post_hooks: Dict[int, Callable]

_load_state_dict_pre_hooks: Dict[int, Callable]

_modules: Dict[str, Optional[Module]]

_non_persistent_buffers_set: Set[str]

_parameters: Dict[str, Optional[Parameter]]

_state_dict_hooks: Dict[int, Callable]

forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

mu(x)

training: bool

init_vae(img_size, latent_dim, loss_f, name)

Return an instance of a VAE with encoder and decoder from model_type.

log_density_gaussian(x: Tensor, mu: Tensor, logvar: Tensor)

Computes the log pdf of the Gaussian with parameters mu and logvar at x

Parameters

• x – (Tensor) Point at whichGaussian PDF is to be evaluated

• mu – (Tensor) Mean of the Gaussian distribution

• logvar – (Tensor) Log variance of the Gaussian distribution