Last week, we first designed our CNN model with four convolution layers.
The CNN model we designed has a dropout value of 0.2 within the convolution layers. As an optimizer, it had the ‘adam’ optimizer.
Our goal this week is to find the most successful general CNN model by changing the number of convolution layers and playing with the optimizer and dropout values.
During this study, we will experiment on 2, 3 and 4-layer models, depending on the process sequence specified below, and choose the most successful model.
- The results we will obtain with the same parameters as our main model (for each)
- The results we will get when we change the optimizer and dropout
This sequence of operations will be followed sequentially, and our aim will be to see how each change affects our new models.
As a result of these operations. If any model does not perform better than our main model, we will not make a model change.
Our basic model: (With four convolutional layers, ’adam’ optimizer and 0.2 Dropout rate.)
2 Layer Model Tests
Adam Optimizer & 0.2 Dropout
Final Loss and MSE Values: 0.0104 – 0.0029
Loss Graph:
First Accuracy:
Last Accuracy:
Parameters:
Adam Optimizer & 0.5 Dropout
Final Loss and MSE Values: 0.0303 – 0.0087
Loss Graph:
First Accuracy:
Last Accuracy:
Parameters:
SGD Optimizer & 0.2 Dropout
Final Loss and MSE Values: 0.0947 – 0.0264
Loss Graph:
First Accuracy:
Last Accuracy:
Parameters:
SGD Optimizer & 0.5 Dropout
Final Loss and MSE Values: 0.1030 – 0.0287
Loss Graph:
First Accuracy:
Last Accuracy:
Parameters:
3 Layer Model Tests
Adam Optimizer & 0.2 Dropout
Final Loss and MSE Values: 0.0199 – 0.0057
Loss Graph:
First Accuracy:
Last Accuracy:
Parameters:
Adam Optimizer & 0.5 Dropout
Final Loss and MSE Values: 0.0431 – 0.0124
Loss Graph:
First Accuracy:
Last Accuracy:
Parameters:
SGD Optimizer & 0.2 Dropout
Final Loss and MSE Values: 0.1038 – 0.0289
Loss Graph:
First Accuracy:
Last Accuracy:
Parameters:
SGD Optimizer & 0.5 Dropout
Final Loss and MSE Values: 0.1218 – 0.0340
Loss Graph:
First Accuracy:
Last Accuracy:
Parameters:
4 Layer Model Tests
Adam Optimizer & 0.2 Dropout
Final Loss and MSE Values: 0.0334 – 0.0096
Loss Graph:
First Accuracy:
Last Accuracy:
Parameters:
Adam Optimizer & 0.5 Dropout
Final Loss and MSE Values: 0.0560 – 0.0160
Loss Graph:
First Accuracy:
Last Accuracy:
Parameters:
SGD Optimizer & 0.2 Dropout
Final Loss and MSE Values: 0.1219 – 0.0336
Loss Graph:
First Accuracy:
Last Accuracy:
Parameters:
SGD Optimizer & 0.5 Dropout
Final Loss and MSE Values: 0.1420 – 0.0387
Loss Graph:
First Accuracy:
Last Accuracy:
Parameters:
