Table 4 Range of hyper-parameter values for deep learning models
From: Social media text analytics of Malayalam–English code-mixed using deep learning
Hyper-parameters | Offensive task | Sentiment task |
|---|---|---|
Learning rate | 10–5 to 1 | 10–5 to 1 |
Dropout | 0 to 1 | 0 to 1 |
Epochs | 5 and 60 with intervals of 10 | 10 and 400, with an interval of 10 |
Word embedding dimensions | 100 to 500 | 100 to 800 |
Batch size | 32, 64, 128, and 256 | 32, 64, 128, 256 and 512 |
Window size | 2 to 10 | 2 to 10 |
Maximum sequence length | 32 and 64 | 8, 10, 16 |
Hidden units | 80 to 150 | 80 to 150 |
Loss function | Cross-entropy | Cross-entropy |
Activation functions | elu, selu, relu | elu, selu, relu |
Optimization Algorithm | Adagrad, Adadelta, RMSprop, Stochastic Gradient Descent(SGD) and Adam | Adagrad, Adadelta, RMSprop, Stochastic Gradient Descent(SGD) and Adam |