machine learning - Confusion between Binary_crossentropy and Categorical_crossentropy -
i doing binary class classification using deep neural network. whenever using binary_crossentropy model not giving accuracy (it closer random prediction). if use categorical crossentropy making size of output layer 2, getting accuracy in 1 epoch close 0.90. can please explain happening here?
i have problem while trying use binary_crossentropy softmax
activation in output layer. far know, softmax
give probability of each class, if output layer has 2 nodes, p(x1)
, p(x2)
, x1 + x2 = x
. therefore, if have 1 output node, equals 1.0 (100%), that's why have close random prediction (honestly, close category distribution in evaluation set).
try changing activation method sigmoid
or relu
.
Comments
Post a Comment