val_acc and top-1 acc are different?

[ganab...@gmail.com]

Hello all

After i finished training my dataset i have noticed that val_top-1 accuracy and val_acc are different.

val_acc and top-1 acc must be same as explained in concept of top-n acc.

Why would it happen?

Ex)

acc: 0.8876 - in_top1: 0.0444

val_acc: 0.8850 - val_in_top1: 0.0222

code:

def top_k_categorical_accuracy(y_true, y_pred, k=5):

    return K.mean(K.in_top_k(y_pred, K.argmax(y_true, axis=-1), k), axis=-1)

def in_top1(x, y):

    return top_k_categorical_accuracy(x, y, k=1)

def in_top5(x, y):

    return top_k_categorical_accuracy(x, y, k=5)

metrics = ['accuracy']

metrics.append(in_top1)

metrics.append(in_top5)

new_model.compile(loss='binary_crossentropy',

              optimizer='rmsprop',

              metrics=metrics)

result:

ETA: 0s - loss: 1.8021 - acc: 0.8876 - in_top1: 0.0444 - in_top2: 1.0000 - in_top3: 1.0000 - in_top4: 1.0000 - in_top5: 1.0000 - in_top6: 1.0000 - in_top7: 1.0000 - in_top8: 1.0000 - in_top9: 1.0000 - in_top10: 1.0000 - val_loss: 1.8440 - val_acc: 0.8850 - val_in_top1: 0.0222 - val_in_top2: 1.0000 - val_in_top3: 1.0000 - val_in_top4: 1.0000 - val_in_top5: 1.0000 - val_in_top6: 1.0000 - val_in_top7: 1.0000 - val_in_top8: 1.0000 - val_in_top9: 1.0000 - val_in_top10: 1.0000