cornet_params

CORNET_SOURCE = 'Biswas et al., IEEE TBioCAS 13(2):282-291, 2019, doi:10.1109/TBCAS.2019.2892297'

CorNetParams()

CorNET heart-rate regressor: two convolution stages then stacked LSTMs.

Defaults are the paper’s HR network (Sec. III, Fig. 6, Table III): two Conv1D(32, 40) stages with batch normalization, ReLU, max pooling 4 and dropout 0.1, then LSTM(128) twice and a single linear output. The paper publishes no code or weights; changed values are new architectures. unroll builds the LSTMs as per-timestep operations instead of a loop; weights are identical.

model_config = ConfigDict(frozen=True, extra='forbid')

family: Literal['cornet'] = 'cornet'

conv_stages: int = Field(default=2, ge=1)

filters: int = Field(default=32, gt=0)

kernel_size: int = Field(default=40, gt=0)

pool_size: int = Field(default=4, gt=0)

dropout: float = Field(default=0.1, ge=0, lt=1, allow_inf_nan=False)

lstm_layers: int = Field(default=2, ge=1)

lstm_units: int = Field(default=128, gt=0)

recurrent_activation: Literal['sigmoid', 'hard_sigmoid'] = 'sigmoid'

unroll: bool = False