Shooting's Blog


Coursera course



  • Various way to adjust the model.

    • Increase dense layer -> Training takes longer, but is more accurate.
    • Remove Flatten() layer -> Get an error about the shape of the data.
    • Revise the final output layer to be differed from classified category -> Get an error as soon as it finds an unexpected value.
    • Add additional layers in the network -> No big impact due to simple data.
    • More poches -> overfitting
    • Don't normalize data
    • Stop the training when reached a desired value
  • Stop the training when it reaches a certain condition, such as accuracy>0.5 or loss<0.5

    • Implement a callback to stop training.
    • Check the condition at the end of epoch.
    • Because while data is not completed processed, the acc or loss may be up and down.
    • Trigger the callback.
  • Exercise sheet