Skip to main content

Table 1 Hyper-parameter values of Tree-LSTM

From: Querying knowledge graphs in natural language

Parameter

Value

Input dimensions

300 × 1

LSTM memory dimensions

150 × 1

Epochs

15

Mini batch size

25

Learning rate

1 × \(10^{-2}\)

Weight decay (Regularization)

2.25 × \(10^{-3}\)

Dropout

0.2

Loss function

Kullback-Leibler divergence loss

Optimizer

Adagrad optimizer

Learning rate scheduler

Stepwise learning rate decay

Step learning rate step size

Once every 2 epochs

Step learning rate decay

0.25