Bert#

class tfts.models.bert.Bert(predict_sequence_length: int = 1, config: BertConfig | None = None)[source]#

Bases: BaseModel

Bert model for time series forecasting.

This model implements a transformer-based architecture (BERT) adapted for time series data. It processes time series inputs through a transformer encoder and produces predictions for future time steps.

Parameters:

  • predict_sequence_length (int, optional) – Number of future time steps to predict, by default 1

  • config (BertConfig, optional) – Configuration parameters for the model, by default None

config#

Configuration object containing model hyperparameters

Type:

BertConfig

predict_sequence_length#

Number of future time steps to predict

Type:

int

encoder_embedding#

Embedding layer for encoder inputs

Type:

DataEmbedding

encoder#

Transformer encoder module

Type:

Encoder

dense_layers#

List of dense layers for final projection

Type:

List[Dense]

Inherited-members:

Methods

build_model(inputs)

compute_output_shape(input_shape)

get_config()

load_pretrained_weights(weights_dir)

predict(x, **kwargs)

save_model(weights_dir)

save_pretrained(save_directory[, ...])

save_weights(weights_path)

summary()

to_model()