Threshold to predict new data

[Hari-pyt]

can i include round() in the final layer

or just put a threshold 0.5 for the sigmoid outputs for new predictions

[marcoaleixo]

how you fixed that?

[Hari-pyt]

Sigmoid units will always give the probability of which class it belongs. So in cross validation ( evaluation) it round up the probability value and got 81 percent accuracy. So I placed a threshold on output file, if probability>0.5 = 1 else 0.

[Hari-pyt]

Also learned that if we add round() in final layer, network struggle with no improvement in accuracy.