In the project, PneumoniaMNIST was used as dataset ,.The following experiments have been done.
- The DNN was defined with six layers,you can find the defination in config.py
Four experiments:
- Base model is used to overfit the dataset without dropout.
- Increasing model is used to avoid overfit with droput in every layer .
- dropout ->[0, 0.1, 0.3, 0.5, 0.7, 0.9]
- Reduce model is used to avoid overfit with droput in every layer.
- dropout ->[0.9, 0.7, 0.5, 0.3, 0.1, 0]
- Avg model uses dropout in every layer, and all the dropout weights are the same.
- dropout ->[0.41,0.41,0.41,0.41,0.41,0.41] 2.5/6 <-->0.41