Skip to content

Latest commit

 

History

History
48 lines (38 loc) · 1.06 KB

Model_Details.md

File metadata and controls

48 lines (38 loc) · 1.06 KB
Raw

Model 1

  • Flatten layer (784)
  • Dense layer (256)(relu)
  • Dense layer (128)(relu)
  • Dropout (0.45)
  • Dense layer (62)(Output)(softmax)

Model 2

  • Conv2D layer (32)(kernel = 5)(relu)
  • MaxPool2D layer
  • Dropout (0.3)
  • Flatten
  • Dense Layer(128)(relu)
  • Dense Layer (62)(output)(softmax)

Model 3

  • Conv2D (32)(kernel = 3)
  • MaxPool2D
  • BatchNormalization()
  • Conv2D (64)(kernel = 3)
  • BatchNormalization()
  • MaxPool2D
  • BatchNormalization()
  • Conv2D (256)(kernel = 3)
  • BatchNormalization()
  • Conv2D (256)(kernel = 3)
  • GlobalAvgPool2D
  • Dense (256)(relu)
  • Dense (62)(softmax)

Future

I am planning on doing a ResNet but the concept seems scary.

PS

I have heard that LSTM RNN is good for this. But again that's scary.

PS PS

I deleted model 4 because it was faulty. IDK why tho and can't be bothered to check.

I am slowly coming to the conclusion that with this dataset I cannot get past the 84% threshold. The conv nets produce weird results in inference despite having better stats.