pe.embedding package

class pe.embedding.CLIP(res=None, device='cuda', batch_size=2000)

Bases: Embedding

Compute the CLIP embedding of images.

__init__(res=None, device='cuda', batch_size=2000)

Constructor.

Parameters:

• res (int, optional) – The resolution of the images. The images will be resized to (res, res) before computing the embedding. If None, the images will not be resized. Defaults to None

• device (str, optional) – The device to use for computing the embedding, defaults to “cuda”

• batch_size (int, optional) – The batch size to use for computing the embedding, defaults to 2000

compute_embedding(data)

Compute the CLIP embedding of images.

Parameters:

data (pe.data.Data) – The data object containing the images

Returns:

The data object with the computed embedding

Return type:

pe.data.Data

class pe.embedding.Embedding

Bases: ABC

The abstract class that computes the embedding of samples.

property column_name

The column name to be used in the data frame.

abstract compute_embedding(data)

Compute the embedding of samples.

Parameters:

data (pe.data.Data) – The data to compute the embedding

filter_uncomputed_rows(data)

merge_computed_rows(data, computed_data)

class pe.embedding.FLDInception(res=None)

Bases: Embedding

Compute the Inception embedding of images using FLD library.

__init__(res=None)

Constructor.

Parameters:

res (int, optional) – The resolution of the images. The images will be resized to (res, res) before computing the embedding. If None, the images will not be resized. Defaults to None

compute_embedding(data)

Compute the Inception embedding of images.

Parameters:

data (pe.data.Data) – The data object containing the images

Returns:

The data object with the computed embedding

Return type:

pe.data.Data

class pe.embedding.Inception(res, device='cuda', batch_size=2000)

Bases: Embedding

Compute the Inception embedding of images.

__init__(res, device='cuda', batch_size=2000)

Constructor.

Parameters:

• res (int) – The resolution of the images. The images will be resized to (res, res) before computing the embedding

• device (str, optional) – The device to use for computing the embedding, defaults to “cuda”

• batch_size (int, optional) – The batch size to use for computing the embedding, defaults to 2000

compute_embedding(data)

Compute the Inception embedding of images.

Parameters:

data (pe.data.Data) – The data object containing the images

Returns:

The data object with the computed embedding

Return type:

pe.data.Data

class pe.embedding.RawPixel

Bases: Embedding

Use the raw pixels of images as the embedding.

compute_embedding(data)

Extract the raw pixels of images.

Parameters:

data (pe.data.Data) – The data object containing the images

Returns:

The data object with the computed embedding

Return type:

pe.data.Data

class pe.embedding.SentenceTransformer(model, batch_size=2000)

Bases: Embedding

Compute the Sentence Transformers embedding of text.

__init__(model, batch_size=2000)

Constructor.

Parameters:

• model (str) – The Sentence Transformers model to use

• batch_size (int, optional) – The batch size to use for computing the embedding, defaults to 2000

property column_name

The column name to be used in the data frame.

compute_embedding(data)

Compute the Sentence Transformers embedding of text.

Parameters:

data (pe.data.Data) – The data object containing the text

Returns:

The data object with the computed embedding

Return type:

pe.data.Data

class pe.embedding.TabularEmbedding(info, cat_weight=0.3333333333333333, num_weight=1)

Bases: Embedding

Compute the tabular embedding.

__init__(info, cat_weight=0.3333333333333333, num_weight=1)

Constructor.

Parameters:

• info (dict) – The information (categories and numerical bounds) of the private data

• cat_weight (float, optional) – The weight for the categorical columns, defaults to 1/3

• num_weight (float, optional) – The weight for the numerical columns, defaults to 1

compute_embedding(data)

Compute the tabular embedding. (the embedding is computed using the features only, not the labels) Vectorization per column is implemented to improve the performance.

Parameters:

data (pe.data.Data) – The data object containing the tabular data

Returns:

The data object with the computed embedding

Return type:

pe.data.Data

Subpackages

• pe.embedding.image package
  • Submodules
    • pe.embedding.image.fld_inception module
      • FLDInception
        • FLDInception.__init__()
        • FLDInception.compute_embedding()
      • to_uint8()
    • pe.embedding.image.inception module
      • Inception
        • Inception.__init__()
        • Inception.compute_embedding()
      • to_uint8()
• pe.embedding.tabular package
  • Submodules
    • pe.embedding.tabular.tabular_embedding module
      • TabularEmbedding
        • TabularEmbedding.__init__()
        • TabularEmbedding.compute_embedding()
• pe.embedding.text package
  • Submodules
    • pe.embedding.text.sentence_transformer module
      • SentenceTransformer
        • SentenceTransformer.__init__()
        • SentenceTransformer.column_name
        • SentenceTransformer.compute_embedding()

Submodules

• pe.embedding.embedding module
  • Embedding
    • Embedding.column_name
    • Embedding.compute_embedding()
    • Embedding.filter_uncomputed_rows()
    • Embedding.merge_computed_rows()