echoflow package

Top-level package for EchoFlow.

class echoflow.EchoFlow(lr: float = 0.0001, nb_epochs: int = 1000, batch_size: int = 100, nb_blocks: int = 3, block_type: str = 'RNVP', use_kde: bool = False)

Bases: object

Wrapper for training normalizing flow models.

fit(df: pandas.core.frame.DataFrame, context: Optional[pandas.core.frame.DataFrame] = None)

Fit the flow model.

Parameters

• df – The dataframe containing the samples to model.

• contexts – The (optional) context dataframe for conditional sampling.

log_likelihood(df: pandas.core.frame.DataFrame, context: Optional[pandas.core.frame.DataFrame] = None)

Compute the log-likelihood of the data.

Parameters

• df – The dataframe containing the samples to model.

• contexts – The (optional) context dataframe. If it was provided in the fit method, it must be provided here as well.

sample(num_samples: Optional[int] = None, context: Optional[pandas.core.frame.DataFrame] = None) → pandas.core.frame.DataFrame

Generate samples via the inverse transform.

Either num_samples or contexts must be provided. If both are provided, then they must be consistent (i.e. there are num_samples rows in contexts).

Parameters

• num_samples – The number of samples.

• contexts – The (optional) context dataframe for conditional sampling.

Subpackages

• echoflow.core package
  • Submodules
  • echoflow.core.base module
  • echoflow.core.batch_norm module
  • echoflow.core.coupling module
  • echoflow.core.made module
  • echoflow.core.one_hot module
  • echoflow.core.reverse module
  • echoflow.core.sequential module
• echoflow.demo package

Submodules

echoflow.echoflow module

class echoflow.echoflow.EchoFlow(lr: float = 0.0001, nb_epochs: int = 1000, batch_size: int = 100, nb_blocks: int = 3, block_type: str = 'RNVP', use_kde: bool = False)

Bases: object

Wrapper for training normalizing flow models.

fit(df: pandas.core.frame.DataFrame, context: Optional[pandas.core.frame.DataFrame] = None)

Fit the flow model.

Parameters

• df – The dataframe containing the samples to model.

• contexts – The (optional) context dataframe for conditional sampling.

log_likelihood(df: pandas.core.frame.DataFrame, context: Optional[pandas.core.frame.DataFrame] = None)

Compute the log-likelihood of the data.

Parameters

• df – The dataframe containing the samples to model.

• contexts – The (optional) context dataframe. If it was provided in the fit method, it must be provided here as well.

sample(num_samples: Optional[int] = None, context: Optional[pandas.core.frame.DataFrame] = None) → pandas.core.frame.DataFrame

Generate samples via the inverse transform.

Either num_samples or contexts must be provided. If both are provided, then they must be consistent (i.e. there are num_samples rows in contexts).

Parameters

• num_samples – The number of samples.

• contexts – The (optional) context dataframe for conditional sampling.

class echoflow.echoflow.FlowDataset(continuous, categorical, contexts)

Bases: torch.utils.data.dataset.Dataset

echoflow.transformer module

class echoflow.transformer.KDETransformer

Bases: object

Probability integral transform via KDE.

fit(X: numpy.ndarray)

inverse_transform(Z: numpy.ndarray, maxiter=1000, tol=1e-06) → numpy.ndarray

transform(X: numpy.ndarray) → numpy.ndarray

class echoflow.transformer.SplitTransformer(use_kde)

Bases: object

fit(df: pandas.core.frame.DataFrame)

fit_transform(df: pandas.core.frame.DataFrame) → Tuple[Optional[torch.Tensor], Optional[torch.Tensor]]

inverse_transform(continuous: Optional[torch.Tensor], categorical: Optional[torch.Tensor]) → pandas.core.frame.DataFrame

transform(df: pandas.core.frame.DataFrame) → Tuple[Optional[torch.Tensor], Optional[torch.Tensor]]

class echoflow.transformer.TableTransformer(use_kde)

Bases: object

Transform a dataframe into a tensor.

dims = 0

fit_transform(df: pandas.core.frame.DataFrame) → torch.Tensor

Fit and transform a dataframe into a tensor.

Continuous values are normalized to the [0.0, 1.0] range and categorical values are converted into a one-hot representation.

Parameters

df – The dataframe containing continuous and categorical values.

Returns

A tensor representation of the data.

Return type

torch.Tensor

inverse_transform(inputs: torch.Tensor) → pandas.core.frame.DataFrame

Inverse transform a tensor into a dataframe.

Parameters

inputs – The tensor to apply the inverse transform to.

transform(df)