kerasで多値分類の評価関数を自作したい

kerasで三値分類の評価関数を自作したい

ここに質問の内容を詳しく書いてください。
y_true = [0,1,0]かつy_pred = [0,1,0]が非常に多いのでIoUのようにして
このような場合を除いたときの精度を求めようとしているのですがうまく行きません。
よろしくお願いします。

該当のソースコード

Python
1model = tensorflow.keras.models.Sequential()
2dense_1 = Dense(30,input_dim=399, activation = 'relu')
3dense_2 = Dense(30, activation = 'relu')
4dense_3 = Dense(3, activation = 'sigmoid')
5model.add(dense_1)
6model.add(dense_2)
7model.add(dense_3)
8model.compile(loss='categorical_crossentropy',optimizer='adam',metrics=['accuracy', IoU])
9model.fit(x_train, y_train, epochs=1, batch_size=1, validation_data=(x_test, y_test))

Python
1#IoU
2def iou0_0(y_true, y_pred):
3    print(y_true[0])
4    print(y_pred[0])
5    return K.sum(K.cast(K.equal(y_true[0] * y_pred[0], 1), K.floatx()))
6def iou0_1(y_true, y_pred):
7    return K.sum(K.cast(K.equal(y_true[0] * y_pred[1], 1), K.floatx()))
8def iou0_2(y_true, y_pred):
9    return K.sum(K.cast(K.equal(y_true[0] * y_pred[2], 1), K.floatx()))
10def iou1_0(y_true, y_pred):
11    return K.sum(K.cast(K.equal(y_true[1] * y_pred[0], 1), K.floatx()))
12def iou1_2(y_true, y_pred):
13    return K.sum(K.cast(K.equal(y_true[1] * y_pred[2], 1), K.floatx()))
14def iou2_0(y_true, y_pred):
15    return K.sum(K.cast(K.equal(y_true[2] * y_pred[0], 1), K.floatx()))
16def iou2_1(y_true, y_pred):
17    return K.sum(K.cast(K.equal(y_true[2] * y_pred[1], 1), K.floatx()))
18def iou2_2(y_true, y_pred):
19    return K.sum(K.cast(K.equal(y_true[2] * y_pred[2], 1), K.floatx()))  
20
21def IoU(y_true, y_pred):
22    y_pred = K.round(y_pred)
23    return (iou0_0(y_true, y_pred) + iou2_2(y_true, y_pred))/ iou0_0(y_true, y_pred)+iou0_1(y_true, y_pred)+iou0_2(y_true, y_pred)+iou1_0(y_true, y_pred)+iou1_2(y_true, y_pred)+iou2_0(y_true, y_pred)+iou2_1(y_true, y_pred)+iou2_2(y_true, y_pred)

発生している問題・エラーメッセージ

　　 402   model.add(dense_3)
    403   model.compile(loss='categorical_crossentropy',optimizer='adam',metrics=['accuracy', IoU])
--> 404   model.fit(x_train, y_train, epochs=1, batch_size=1, validation_data=(x_test, y_test))

    406 def main():


    725         max_queue_size=max_queue_size,
    726         workers=workers,
--> 727         use_multiprocessing=use_multiprocessing)
    728 
    729   def evaluate(self,


    673         validation_steps=validation_steps,
    674         validation_freq=validation_freq,
--> 675         steps_name='steps_per_epoch')
    676 
    677   def evaluate(self,

    392 
    393         # Get outputs.
--> 394         batch_outs = f(ins_batch)
    395         if not isinstance(batch_outs, list):
    396           batch_outs = [batch_outs]


   3474 
   3475     fetched = self._callable_fn(*array_vals,
-> 3476                                 run_metadata=self.run_metadata)
   3477     self._call_fetch_callbacks(fetched[-len(self._fetches):])
   3478     output_structure = nest.pack_sequence_as(


   1470         ret = tf_session.TF_SessionRunCallable(self._session._session,
   1471                                                self._handle, args,
-> 1472                                                run_metadata_ptr)
   1473         if run_metadata:
   1474           proto_data = tf_session.TF_GetBuffer(run_metadata_ptr)

InvalidArgumentError: slice index 2 of dimension 0 out of bounds.
	 [[{{node metrics_46/IoU/strided_slice_20}}]]