出現NoneType' object is not subscriptable錯誤 - Cupoy

2019/12/07 上午 11:52

機器學習共學討論版

骑士

觀看數：107

回答數：2

收藏數：0

ml100-3

ml100-3-d88

TypeError                                 Traceback (most recent call last)
<ipython-input-8-de2a5ca0ea56> in <module>
     9           validation_data=(x_test, y_test),
    10           shuffle=True,
---> 11           callbacks=[log_f1sc]
    12          )
    13 

....

16     def on_epoch_end(self, epoch, logs = {}, thres=0.5):
    17         logs = logs or {}
---> 18         y_true = self.validation_data[1].argmax(axis = 1)
    19         y_pred = self.model.predict(self.validation_data[0])
    20         y_pred = (y_pred[:, 1] >= thres) * 1

TypeError: 'NoneType' object is not subscriptable

备注：我是调用tensorflow.keras函数库。

回答列表

• 2019/12/11 下午 11:08
  Seanyu.TW
  贊同數：0
  不贊同數：0
  留言數：0

  Hi, 因為你的 self.validation_data[1] 是 None, 建議你在函式裡面 print(self.validation_data[1])

• 2020/07/31 下午 07:50
  Shengzhi Lin
  贊同數：0
  不贊同數：0
  留言數：0

  嗨，我發現在callback函數裡面沒辦法直接繼承model validation data，所以要在寫class的時候加進去，我的code如下所示：

  
  

  class f1sc(keras.callbacks.Callback): def __init__(self, val_data=(x_test, y_test)): self.validation_data = val_data def on_train_begin(self, epoch, logs = {}): logs = logs or {} record_items = ["val_auc", "val_f1sc", "val_fp", "val_fn", "val_tp", "val_tn"] for i in record_items: try: if i not in self.params['metrics']: self.params['metrics'].append(i) except: self.params['metrics']=i def on_epoch_end(self, epoch, logs = {}, thres=0.5): logs = logs or {} y_true = self.validation_data[1].argmax(axis = 1) y_pred = self.model.predict(self.validation_data[0]) y_pred = (y_pred[:, 1] >= thres) * 1 logs["val_f1sc"] = f1_score(y_true = y_true, y_pred = y_pred, average="weighted")