regression_2_loss.py

def missing_loss(y_true, y_pred):

    # y_pred=tf.constant([[1.0,2.0],[5.0,10.0]])



    bool_finite = tf.where(tf.equal(y_true, -1.0))
    int_finite = K.cast(bool_finite, dtype='int32')
    float_finite = K.cast(int_finite, dtype='float32')
    y_pred = K.cast(y_pred, dtype='float32')


    #bool_finite_val = bool_finite.eval()
    #bool_finite_val =bool_finite_val.astype(np.float32)
    #new_bool_finite = tf.constant(bool_finite_val)

    y_true = K.print_tensor(y_true, message='y_true = ')
    y_pred = K.print_tensor(y_pred, message='y_pred = ')

    #sea, fresh, total, ps, pf, pt = y_true
    # hvor pX er 1 dersom vi har en verdi for X
    #pred_sea, pred_tot... = y_pred

    loss = K.square( y_pred  - y_true )
    loss = K.tf.multiply( loss, float_finite)
    loss = K.sum(loss)
    #K.print_tensor(loss, message='loss=')
    return loss