Skip to main content

Table 1 LSTM Network configurations

From: Prediction and detection of freezing of gait in Parkinson’s disease from plantar pressure data using long short-term memory neural-networks

Hyperparameter Values tested
Number of LSTM layers 1, 2, 3, 4, 5
Number of units in each LSTM layer 16, 32, 64
Constant learning rate 0.1, 0.01, 0.001, 0.0001
Learning rate decay with a decay rate
(decay rate, initial learning rate)
(0.5, 0.005), (0.75, 0.001)
Learning rate decreases in discrete steps
(initial learning rate)
Decreases to half every 5 epochs (0.01)