From: Application of deep learning technique in next generation sequence experiments
Parameters | ART simulation data | Chr 22 WGS data |
---|---|---|
Drop-out | 0.2 (800 train set/200 test set) | 0.2 (360 train set/88 test set) |
Deep learning | ||
Batch size | 160 | 60 |
Epoch | 500; 1000; 2000 | 500; 1000; 2000 |
Number of iterations for each epoch in the model | 5 | 6 |
Iteration | 10,000 | 12,000 |
Learning rate (LR) | 0.01; 0.001 | 0.01; 0.001 |
LightGBM | ||
Min data in leaf | 100 | 100 |
Max depth | 7 | 5 |
Num leaves | 128 | 32 |
Num iterations | 100 | 100 |
Learning rate (LR) | 0.01 | 0.001 |
Bagging fraction | 0.5 | 0.5 |
XGBoost | ||
Eta | 0.01 | 0.015 |
Min child weight | 1.4 | 1 |
Max depth | 5 | 3 |
Gamma | 0.1 | 0.1 |
Alpha | 0.001 | 0.001 |
Lambda | 1 | 1 |
Subsample | 0.8 | 0.8 |