Table 3 The parameter design of the model.
From: A hybrid model combining 1D-CNN and BERT for intelligent ECG arrhythmia classification
Parameter | Description |
|---|---|
num_inputs=1 | Input dimension at each time step (usually single-channel for ECG) |
num_channels=[1,1,1] | Number of output channels in each TCN layer (affects feature extraction capacity) |
kernel_size=5 | Kernel size of the TCN (determines the temporal window size) |
dropout1=0.3 | Dropout rate in the TCN module (prevents overfitting) |
d_feature | Length of the input signal |
d_model | Embedding dimension in the Transformer (key hyperparameter) |
d_inner | Intermediate dimension of the Feed-Forward Network (FFN) |
n_layers | Number of Transformer layers |
d_layers | Number of TCN layers (possibly customized) |
n_head | Number of attention heads in the multi-head attention |
d_k, d_v=64 | Dimensions of keys and values in the attention mechanism |
dropout | Overall dropout rate in the Transformer |
class_num | Number of output classes (i.e., types of heart rhythms) |
