Question

我正在制作一个预测3个班级的CNN。如果我将批次大小设置得足够小（有时甚至会更大），则会收到类似自定义ROC指标的错误（请参见下面的功能和错误）。我看过另一个关于我的错误的stackoverflow问题，但没有解决在keras度量中使用它的问题。我的问题是有没有办法使用try，除了通过绕过错误并进入下一个时期而不终止训练？我知道解决此问题的实际方法是拥有更平衡的数据，但是我仍然想知道当前设置中的问题是否有解决方案。谢谢。顺便说一句，即使我强制数据生成器在数据集中至少包含2个类，我仍然似乎会收到此错误。

AUC指标

def sk_auroc(y_true, y_pred):
    import tensorflow as tf
    from sklearn.metrics import roc_auc_score
    return tf.py_func(roc_auc_score, (y_true, y_pred), tf.double)

错误

Epoch 12/300
 88/103 [========================>.....] - ETA: 3s - loss: 0.5235 - acc: 0.7960 - sk_auroc: 0.9289 - auc: 0.9058
ValueError: Only one class present in y_true. ROC AUC score is not defined in that case.
    Traceback (most recent call last):

我尝试过

def sk_auroc(y_true, y_pred):
    import tensorflow as tf
    from sklearn.metrics import roc_auc_score

    try:
        roc_auc_score(y_true, y_pred)
    except:
        pass

但是当我尝试编译模型时出现以下错误。据我所知，我需要return

AttributeError                            Traceback (most recent call last)
<ipython-input-152-12b92143dd11> in <module>()
      4 
      5 if t_space['which_model']=='mike':
----> 6     model = build_compile_opo_ai_model_mike(t_space)
      7 
      8 elif t_space['which_model']=='liyuan':

<ipython-input-150-3e0af212444a> in build_compile_opo_ai_model_mike(space)
     79     model.compile(optimizer=Adam(lr=space['learning_rate']*space['lr_rate_mult']),
     80                  loss=space['loss'],
---> 81                  metrics=[space['metrics'], sk_auroc])
     82     print('Plotting model...')
     83     plot_model(model, to_file='mike_model.png')

~/anaconda3/lib/python3.6/site-packages/keras/engine/training.py in compile(self, optimizer, loss, metrics, loss_weights, sample_weight_mode, weighted_metrics, target_tensors, **kwargs)
    449                 output_metrics = nested_metrics[i]
    450                 output_weighted_metrics = nested_weighted_metrics[i]
--> 451                 handle_metrics(output_metrics)
    452                 handle_metrics(output_weighted_metrics, weights=weights)
    453 

~/anaconda3/lib/python3.6/site-packages/keras/engine/training.py in handle_metrics(metrics, weights)
    418                     metric_result = weighted_metric_fn(y_true, y_pred,
    419                                                        weights=weights,
--> 420                                                        mask=masks[i])
    421 
    422                 # Append to self.metrics_names, self.metric_tensors,

~/anaconda3/lib/python3.6/site-packages/keras/engine/training_utils.py in weighted(y_true, y_pred, weights, mask)
    421             score_array *= weights
    422             score_array /= K.mean(K.cast(K.not_equal(weights, 0), K.floatx()))
--> 423         return K.mean(score_array)
    424     return weighted
    425 

~/anaconda3/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py in mean(x, axis, keepdims)
   1394         A tensor with the mean of elements of `x`.
   1395     """
-> 1396     if x.dtype.base_dtype == tf.bool:
   1397         x = tf.cast(x, floatx())
   1398     return tf.reduce_mean(x, axis, keepdims)

AttributeError: 'NoneType' object has no attribute 'dtype'

如果我像这样加回去，就不会收到上面的错误，但是我仍然会收到原来的错误。我认为除了代码的不同部分外，我需要使用try和，但是我不确定现在在哪里（每历元的度量计算完成了）

def sk_auroc(y_true, y_pred):
    import tensorflow as tf
    from sklearn.metrics import roc_auc_score

    try:
        return tf.py_func(roc_auc_score, (y_true, y_pred), tf.double)
    except:
        pass

除了传递keras自定义指标外，是否可以使用try方法？

0 个答案: