用于 Scikit Learn 的 Keras Wrappers - AUC 记分器不工作

Posted 2023-03-12

【中文标题】用于 Scikit Learn 的 Keras Wrappers - AUC 记分器不工作

【英文标题】：Keras Wrappers for Scikit Learn - AUC scorer is not working

【发布时间】：2016-09-28 04:16:46

【问题描述】：

我正在尝试使用Keras Scikit Learn Wrapper，以便更轻松地随机搜索参数。我在这里写了一个示例代码：

我生成了一个人工数据集：

我正在使用来自scikit learn 的moons

from sklearn.datasets import make_moons
dataset = make_moons(1000)

模型构建器定义：

我定义了build_fn需要的函数：

def build_fn(nr_of_layers = 2,
             first_layer_size = 10,
             layers_slope_coeff = 0.8,
             dropout = 0.5,
             activation = "relu",
             weight_l2 = 0.01,
             act_l2 = 0.01,
             input_dim = 2):

    result_model = Sequential()
    result_model.add(Dense(first_layer_size,
                           input_dim = input_dim,
                           activation=activation,
                           W_regularizer= l2(weight_l2),
                           activity_regularizer=activity_l2(act_l2)
                           ))

    current_layer_size = int(first_layer_size * layers_slope_coeff) + 1

    for index_of_layer in range(nr_of_layers - 1):

        result_model.add(BatchNormalization())
        result_model.add(Dropout(dropout))
        result_model.add(Dense(current_layer_size,
                               W_regularizer= l2(weight_l2),
                               activation=activation,
                               activity_regularizer=activity_l2(act_l2)
                               ))

        current_layer_size = int(current_layer_size * layers_slope_coeff) + 1

    result_model.add(Dense(1,
                           activation = "sigmoid",
                           W_regularizer = l2(weight_l2)))

    result_model.compile(optimizer="rmsprop", metrics = ["accuracy"], loss = "binary_crossentropy")

    return result_model

NeuralNet = KerasClassifier(build_fn)

参数网格定义：

然后我定义了一个参数网格：

param_grid = 
    "nr_of_layers" : [2, 3, 4, 5],
    "first_layer_size" : [5, 10, 15],
    "layers_slope_coeff" : [0.4, 0.6, 0.8],
    "dropout" : [0.3, 0.5, 0.8],
    "weight_l2" : [0.01, 0.001, 0.0001],
    "verbose" : [0],
    "batch_size" : [1],
    "nb_epoch" : [30]

RandomizedSearchCV 阶段：

我定义了RandomizedSearchCV 对象并拟合了来自人工数据集的值：

random_search = RandomizedSearchCV(NeuralNet, 
    param_distributions=param_grid, verbose=2, n_iter=1, scoring="roc_auc")
random_search.fit(dataset[0], dataset[1])

我得到的（在控制台中运行此代码后）是：

Traceback (most recent call last):
  File "C:\Anaconda2\lib\site-packages\IPython\core\interactiveshell.py", line 2885, in run_code
    exec(code_obj, self.user_global_ns, self.user_ns)
  File "<ipython-input-3-c5bdbc2770b7>", line 2, in <module>
    random_search.fit(dataset[0], dataset[1])
  File "C:\Anaconda2\lib\site-packages\sklearn\grid_search.py", line 996, in fit
    return self._fit(X, y, sampled_params)
  File "C:\Anaconda2\lib\site-packages\sklearn\grid_search.py", line 553, in _fit
    for parameters in parameter_iterable
  File "C:\Anaconda2\lib\site-packages\sklearn\externals\joblib\parallel.py", line 800, in __call__
    while self.dispatch_one_batch(iterator):
  File "C:\Anaconda2\lib\site-packages\sklearn\externals\joblib\parallel.py", line 658, in dispatch_one_batch
    self._dispatch(tasks)
  File "C:\Anaconda2\lib\site-packages\sklearn\externals\joblib\parallel.py", line 566, in _dispatch
    job = ImmediateComputeBatch(batch)
  File "C:\Anaconda2\lib\site-packages\sklearn\externals\joblib\parallel.py", line 180, in __init__
    self.results = batch()
  File "C:\Anaconda2\lib\site-packages\sklearn\externals\joblib\parallel.py", line 72, in __call__
    return [func(*args, **kwargs) for func, args, kwargs in self.items]
  File "C:\Anaconda2\lib\site-packages\sklearn\cross_validation.py", line 1550, in _fit_and_score
    test_score = _score(estimator, X_test, y_test, scorer)
  File "C:\Anaconda2\lib\site-packages\sklearn\cross_validation.py", line 1606, in _score
    score = scorer(estimator, X_test, y_test)
  File "C:\Anaconda2\lib\site-packages\sklearn\metrics\scorer.py", line 175, in __call__
    y_pred = y_pred[:, 1]
IndexError: index 1 is out of bounds for axis 1 with size 1

当我使用 accuracy 度量而不是使用 scoring = "roc_auc" 时，此代码可以正常工作。谁能解释我怎么了？有没有人有类似的问题？

【参考方案1】：

KerasClassifier 中存在导致此问题的错误。我在 repo 上为它打开了一个问题。 https://github.com/fchollet/keras/issues/2864

修复也在那里。您可以同时定义自己的 KerasClassifier 作为临时解决方法。

class FixedKerasClassifier(KerasClassifier):
    def predict_proba(self, X, **kwargs):
        kwargs = self.filter_sk_params(Sequential.predict_proba, kwargs)
        probs = self.model.predict_proba(X, **kwargs)
        if(probs.shape[1] == 1):
            probs = np.hstack([1-probs,probs]) 
        return probs

【讨论】：

非常感谢您 - 您的解决方案做得很好。干杯:)

优秀的解决方案！