import keras.backend as K
def std_mae(std=1):
def mae(y_true, y_pred):
return K.mean(K.abs(y_pred - y_true)) * std
return mae
def std_rmse(std=1):
def rmse(y_true, y_pred):
return K.sqrt(K.mean(K.square((y_pred - y_true)))) * std
return rmse
def std_r2(std=1):
def r2(y_true, y_pred):
ss_res = K.sum(K.square((y_true - y_pred) * std))
ss_tot = K.sum(K.square((y_true - K.mean(y_true) * std)))
return 1 - ss_res / (ss_tot + K.epsilon())
return r2