TypeError: ratio()缺少1个必需的位置参数：“metric_fun”

Question

我正在尝试使用ibm的aif360库来进行去偏。我正在研究一个线性回归模型，并希望尝试一个度量来计算特权组和非特权组之间的差异。但是，当运行这段代码时，我得到以下错误：

TypeError: difference()缺少1个必需的位置参数：“metric_fun”

我已经研究了这个函数的类，但它们引用了一个metric_fun，也读取了文档，但没有得到进一步的信息。该函数缺少一个参数，但我不知道它需要哪个参数。

代码的一小段是：

train_pp_bld = StructuredDataset(df=pd.concat((x_train, y_train),
                                                axis=1),
                                  label_names=['decile_score'],
                                  protected_attribute_names=['sex_Male'],
                                  privileged_protected_attributes=1,
                                  unprivileged_protected_attributes=0)

privileged_groups = [{'sex_Male': 1}]
unprivileged_groups = [{'sex_Male': 0}]

# Create the metric object
metric_train_bld = DatasetMetric(train_pp_bld,
                                            unprivileged_groups=unprivileged_groups,
                                            privileged_groups=privileged_groups)

# Metric for the original dataset
metric_orig_train = DatasetMetric(train_pp_bld, 
                                              unprivileged_groups=unprivileged_groups,
                                              privileged_groups=privileged_groups)
display(Markdown("#### Original training dataset"))
print("Difference in mean outcomes between unprivileged and privileged groups = %f" % metric_orig_train.difference())

给定的堆栈跟踪为：

Traceback (most recent call last):

  File "/Users/sef/Desktop/Thesis/Python Projects/Stats/COMPAS_Debias_AIF360_Continuous_Variable.py", line 116, in <module>
    print("Difference in mean outcomes between unprivileged and privileged groups = %f" % metric_orig_train.difference())

  File "/Users/sef/opt/anaconda3/envs/AI/lib/python3.8/site-packages/aif360/metrics/metric.py", line 37, in wrapper
    result = func(*args, **kwargs)

TypeError: difference() missing 1 required positional argument: 'metric_fun'

创建函数后：

def privileged_value(self, privileged=False):
    if privileged:
        return unprivileged_groups['sex_Male']
    else:
        return privileged_groups['sex_Male']

display(Markdown("#### Original training dataset"))
print("Difference in mean outcomes between unprivileged and privileged groups = %f" % metric_orig_train.difference(privileged_value))

仍然会得到类似的错误回溯：

Traceback (most recent call last):

  File "/Users/sef/Desktop/Thesis/Python Projects/Stats/COMPAS_Debias_AIF360_Continuous_Variable.py", line 123, in <module>
    print("Difference in mean outcomes between unprivileged and privileged groups = %f" % metric_orig_train.difference(privileged_value))

  File "/Users/sef/opt/anaconda3/envs/AI/lib/python3.8/site-packages/aif360/metrics/metric.py", line 37, in wrapper
    result = func(*args, **kwargs)

  File "/Users/sef/opt/anaconda3/envs/AI/lib/python3.8/site-packages/aif360/metrics/dataset_metric.py", line 77, in difference
    return metric_fun(privileged=False) - metric_fun(privileged=True)

  File "/Users/youssefennali/Desktop/Thesis/Python Projects/Stats/COMPAS_Debias_AIF360_Continuous_Variable.py", line 120, in privileged_value
    return privileged_groups['sex_Male']

TypeError: list indices must be integers or slices, not str

有人能给我指个方向吗？在线上没有类似代码的示例。

致以敬意，

Sef

Answer 1

查看GitHub上的库的源代码，需要向difference(self, metric_fun)传递对函数的引用。唯一的区别是将privileged=False作为输入的函数的输出与privileged=True作为输入的函数的输出相减。

def difference(self, metric_fun):
    """Compute difference of the metric for unprivileged and privileged
    groups.
    """
    return metric_fun(privileged=False) - metric_fun(privileged=True)

创建一个类似这样的函数，并将其传递给difference。

def privilege_value(privileged=False) -> int:
    if privileged:
        return unprivileged_groups[0]['sex_male']
    else:
        return privileged_groups[0]['sex_male']

metric_orig_train.difference(privilege_value)

Answer 2

在不了解您正在使用的库的情况下，错误消息看起来仍然非常清楚，特别是因为您只调用了difference一次，如下所示：

metric_orig_train.difference()

错误消息告诉您应该在此调用中传递一个参数。该参数的名称是metric_fun，这向我建议您应该向它传递一个函数引用。

注意: difference()有可能是在您的代码外部调用的。当您提供错误消息时，请始终提交随附的堆栈跟踪(如果有)。然后，我们可以确切地看到问题发生在代码中的哪个位置。