parsanlp package

Submodules

parsanlp.helper_functions module

parsanlp.helper_functions.create_tensorboard_callback(log_dir, name)

parsanlp.helper_functions.unzip_data(filename)

parsanlp.transformers module

class parsanlp.transformers.CrossAttention(*args, **kwargs)

Bases: _BaseAttention

class parsanlp.transformers.Decoder(*args, **kwargs)

Bases: Layer

Decoder module of the transformer model.

Parameters:

• vocab_size (int) – Vocabulary size.

• Nx (int) – Number of decoder layers.

• dmodel (int) – Model dimensionality.

• nheads (int) – Number of attention heads.

• dff (int) – Number of hidden units in the feedforward layer.

• dropout_rate (float, optional) – Dropout rate. Defaults to 0.1.

class parsanlp.transformers.DecoderLayer(*args, **kwargs)

Bases: Layer

Decoder layer of the transformer model.

Parameters:

• dmodel (int) – Model dimensionality.

• nheads (int) – Number of attention heads.

• dff (int) – Number of hidden units in the feedforward layer.

• dropout_rate (float, optional) – Dropout rate. Defaults to 0.1.

class parsanlp.transformers.Encoder(*args, **kwargs)

Bases: Layer

Encoder module of the transformer model.

Parameters:

• vocab_size (int) – Vocabulary size.

• Nx (int) – Number of encoder layers.

• dmodel (int) – Model dimensionality.

• nheads (int) – Number of attention heads.

• dff (int) – Number of hidden units in the feedforward layer.

• dropout_rate (float, optional) – Dropout rate. Defaults to 0.1.

class parsanlp.transformers.EncoderLayer(*args, **kwargs)

Bases: Layer

Encoder layer of the transformer model.

Parameters:

• dmodel (int) – Model dimensionality.

• nheads (int) – Number of attention heads.

• dff (int) – Number of hidden units in the feedforward layer.

• dropout_rate (float, optional) – Dropout rate. Defaults to 0.1.

class parsanlp.transformers.FeedForward(*args, **kwargs)

Bases: Layer

Feedforward neural network layer.

Parameters:

• dff (int) – Number of hidden units in the feedforward layer.

• dmodel (int) – Model dimensionality.

• dropout_rate (float, optional) – Dropout rate. Defaults to 0.1.

class parsanlp.transformers.PositionalEmbedding(*args, **kwargs)

Bases: Layer

Positional embedding layer for transformer models.

Parameters:

• vocab_size (int) – Vocabulary size.

• dmodel (int) – Model dimensionality. Aka embedding dims.

class parsanlp.transformers.SelfAttention(*args, **kwargs)

Bases: _BaseAttention

class parsanlp.transformers.Transformer(*args, **kwargs)

Bases: Layer

Transformer model.

Parameters:

• vocab_size (int) – Vocabulary size.

• output_dims (int) – Dimensionality of the output.

• Nx (int) – Number of encoder and decoder layers.

• dmodel (int) – Model dimensionality.

• nheads (int) – Number of attention heads.

• dff (int) – Number of hidden units in the feedforward layer.

• dropout_rate (float, optional) – Dropout rate. Defaults to 0.1.

parsanlp.transformers.positional_encoding(length, depth)

Generate positional encodings for sequences.

Parameters:

• length (int) – Length of the sequence.

• depth (int) – Depth of the positional encoding.

Returns:

The positional encoding matrix.

Return type:

tf.Tensor

Module contents