sklearn 转换管道和功能联合

Posted

技术标签:

【中文标题】sklearn 转换管道和功能联合【英文标题】:sklearn transformation pipeline and featureunion 【发布时间】:2018-02-16 18:01:09 【问题描述】:

我在尝试运行以下代码时遇到问题。这是房价的机器学习问题。

from sklearn.pipeline import FeatureUnion
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
from sklearn.base import BaseEstimator,TransformerMixin

num_attributes=list(housing_num)
cat_attributes=['ocean_proximity']
rooms_ix, bedrooms_ix, population_ix, household_ix = 3, 4, 5, 6

class DataFrameSelector(BaseEstimator,TransformerMixin):
    def __init__(self,attribute_names):
        self.attribute_names=attribute_names
    def fit(self,X,y=None):
        return self
    def transform(self,X,y=None):
        return X[self.attribute_names].values

class CombinedAttributesAdder(BaseEstimator, TransformerMixin):
    def __init__(self, add_bedrooms_per_room = True): # no *args or **kargs
        self.add_bedrooms_per_room = add_bedrooms_per_room 
    def fit(self, X,y=None):
        return self # nothing else to do 
    def transform(self, X,y=None):
        rooms_per_household = X[:, rooms_ix] / X[:, household_ix] 
        population_per_household = X[:, population_ix] / X[:, household_ix] 
        if self.add_bedrooms_per_room:
            bedrooms_per_room = X[:, bedrooms_ix] / X[:, rooms_ix] 
            return np.c_[X, rooms_per_household, population_per_household, bedrooms_per_room]
        else:
            return np.c_[X, rooms_per_household, population_per_household]


num_pipeline=Pipeline([
    ('selector',DataFrameSelector(num_attributes)),
    ('imputer',Imputer(strategy="median")),
    ('attribs_adder',CombinedAttributesAdder()),
    ('std_scalar',StandardScaler()),
    ])
cat_pipeline=Pipeline([
    ('selector',DataFrameSelector(cat_attributes)),
    ('label_binarizer',LabelBinarizer()),
    ])
full_pipeline=FeatureUnion(transformer_list=[
    ("num_pipeline",num_pipeline),
    ("cat_pipeline",cat_pipeline),
    ])

当我尝试运行时出现错误:

housing_prepared = full_pipeline.fit_transform(housing)

并且错误显示为:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-141-acd0fd68117b> in <module>()
----> 1 housing_prepared = full_pipeline.fit_transform(housing)

/Users/nieguangtao/ml/env_1/lib/python2.7/site-packages/sklearn/pipeline.pyc in fit_transform(self, X, y, **fit_params)
    744             delayed(_fit_transform_one)(trans, weight, X, y,
    745                                         **fit_params)
--> 746             for name, trans, weight in self._iter())
    747 
    748         if not result:

/Users/nieguangtao/ml/env_1/lib/python2.7/site-packages/sklearn/externals/joblib/parallel.pyc in __call__(self, iterable)
    777             # was dispatched. In particular this covers the edge
    778             # case of Parallel used with an exhausted iterator.
--> 779             while self.dispatch_one_batch(iterator):
    780                 self._iterating = True
    781             else:

/Users/nieguangtao/ml/env_1/lib/python2.7/site-packages/sklearn/externals/joblib/parallel.pyc in dispatch_one_batch(self, iterator)
    623                 return False
    624             else:
--> 625                 self._dispatch(tasks)
    626                 return True
    627 

/Users/nieguangtao/ml/env_1/lib/python2.7/site-packages/sklearn/externals/joblib/parallel.pyc in _dispatch(self, batch)
    586         dispatch_timestamp = time.time()
    587         cb = BatchCompletionCallBack(dispatch_timestamp, len(batch), self)
--> 588         job = self._backend.apply_async(batch, callback=cb)
    589         self._jobs.append(job)
    590 

/Users/nieguangtao/ml/env_1/lib/python2.7/site-packages/sklearn/externals/joblib/_parallel_backends.pyc in apply_async(self, func, callback)
    109     def apply_async(self, func, callback=None):
    110         """Schedule a func to be run"""
--> 111         result = ImmediateResult(func)
    112         if callback:
    113             callback(result)

/Users/nieguangtao/ml/env_1/lib/python2.7/site-packages/sklearn/externals/joblib/_parallel_backends.pyc in __init__(self, batch)
    330         # Don't delay the application, to avoid keeping the input
    331         # arguments in memory
--> 332         self.results = batch()
    333 
    334     def get(self):

/Users/nieguangtao/ml/env_1/lib/python2.7/site-packages/sklearn/externals/joblib/parallel.pyc in __call__(self)
    129 
    130     def __call__(self):
--> 131         return [func(*args, **kwargs) for func, args, kwargs in self.items]
    132 
    133     def __len__(self):

/Users/nieguangtao/ml/env_1/lib/python2.7/site-packages/sklearn/pipeline.pyc in _fit_transform_one(transformer, weight, X, y, **fit_params)
    587                        **fit_params):
    588     if hasattr(transformer, 'fit_transform'):
--> 589         res = transformer.fit_transform(X, y, **fit_params)
    590     else:
    591         res = transformer.fit(X, y, **fit_params).transform(X)

/Users/nieguangtao/ml/env_1/lib/python2.7/site-packages/sklearn/pipeline.pyc in fit_transform(self, X, y, **fit_params)
    290         Xt, fit_params = self._fit(X, y, **fit_params)
    291         if hasattr(last_step, 'fit_transform'):
--> 292             return last_step.fit_transform(Xt, y, **fit_params)
    293         elif last_step is None:
    294             return Xt

TypeError: fit_transform() takes exactly 2 arguments (3 given)

所以我的第一个问题是什么导致了这个错误?

得到这个错误后,我试图找出原因,所以我一个一个地运行上面的转换器:

DFS=DataFrameSelector(num_attributes)
a1=DFS.fit_transform(housing)
imputer=Imputer(strategy='median')
a2=imputer.fit_transform(a1)
CAA=CombinedAttributesAdder()
a3=CAA.fit_transform(a2)
SS=StandardScaler()
a4=SS.fit_transform(a3)

DFS2=DataFrameSelector(cat_attributes)
b1=DFS2.fit_transform(housing)
LB=LabelBinarizer()
b2=LB.fit_transform(b1)

result=np.concatenate((a4,b2),axis=1)

这些可以正确执行,除了我得到的 result 是一个大小为 (16512, 16) 的 numpy.ndarray 而housing_prepared = full_pipeline.fit_transform(housing) 的预期结果应该是一个大小为 ( 16512,17)。 所以这是我的第二个问题为什么会导致差异?

Housing 是一个大小为 (16512, 9) 的 DataFrame,只有 1 个分类特征和 8 个数值特征。

提前谢谢你。

【问题讨论】:

第一个错误是由于LabelBinarizer。它只需要一个输入 y,但由于管道,X 和 y 都将发送给它。请分享数据,我可以提供帮助。 @VivekKumar 这是链接,是房屋数据:drive.google.com/file/d/0B12I2_fMO94pVHZhQlVrSlFtZEk/… 为什么你认为结果应该有 17 列而不是 16 列? @VivekKumar 其实我也认为应该是 16 列。但这实际上是教科书上的一个例子。代码是他们的。他们可以成功运行我无法运行的代码,并得到我无法理解的 17 列结果。 【参考方案1】:

看起来 sklearn 以不同于您预期的方式识别数据类型。确保将数字标识为 int。最简单的方法:使用“您的”发布编码的作者提供的数据。 Aurelien Geron Hands on Machine Learning

【讨论】:

【参考方案2】:

我在阅读这本书时遇到了这个问题。在尝试了一堆变通方法(我觉得这是在浪费我的时间)之后,我放弃并安装了 scikit-learn v0.20 dev。下载***here 并使用 pip 安装它。这应该允许您使用为处理这些问题而设计的 CategoricalEncoder 类。

【讨论】:

【参考方案3】:

我遇到了同样的问题,它是由不会总是抛出错误的缩进问题引起的(请参阅https://***.com/a/14046894/3665886)。

如果您直接从书中复制代码,请确保代码正确缩进。

【讨论】:

【参考方案4】:
    TypeError: fit_transform() 只需要 2 个参数(给定 3 个)

为什么会出现这个错误?

答案:因为您使用的是 LabelBinarizer(),即 ideally suitable 作为响应变量。

怎么办?:你有一个few options:

改用 OneHotEncoder() 为 LabelBinarizer 编写自定义转换器 使用支持您的代码的旧版本 sklean
    housing_prepared 的形状不同

如果您使用的是this data,那么您有 9 个预测变量(8 个数字变量和 1 个分类变量)。 CombinedAttributesAdder() 增加了 3 列,LabelBinarizer() 增加了 5 列,所以变成了 17 列 请记住,sklearn.pipeline.FeatureUnion 连接多个转换器对象的结果

当您手动执行此操作时,您不会添加原始的“ocean_proximity”变量。

让我们看看它的实际效果:

print("housing_shape: ", housing.shape)

num_attribs = list(housing_num)
cat_attribs = ["ocean_proximity"]

DFS=DataFrameSelector(num_attribs)
a1=DFS.fit_transform(housing)

print('Numerical variables_shape: ', a1.shape)

imputer=SimpleImputer(strategy='median')
a2=imputer.fit_transform(a1)

a2.shape 

与 a1.shape 相同

CAA=CombinedAttributesAdder()
a3=CAA.fit_transform(a2)
SS=StandardScaler()
a4=SS.fit_transform(a3) # added 3 variables
print('Numerical variable shape after CAA: ', a4.shape, '\n')

DFS2=DataFrameSelector(cat_attribs)
b1=DFS2.fit_transform(housing)

print("Categorical variables_shape: ", b1.shape)

LB=LabelBinarizer()
b2=LB.fit_transform(b1) # instead of one column now we have 5 columns
print('categorical variable shape after LabelBinarization: ', b2.shape) 

增加了 4 列

print(b2)

result=np.concatenate((a4,b2),axis=1)
print('final shape: ', result.shape, '\n') # Final shape

注意:转换列(a4 的结果)和二值化列(b2 的结果)尚未添加到原始数据帧。 为此,您需要将 numpy 数组 b2 转换为数据框

new_features = pd.DataFrame(a4)
new_features.shape

ocean_cat = ['<1H OCEAN', 'INLAND', 'NEAR OCEAN', 'NEAR BAY', 'ISLAND']
ocean_LabelBinarize = pd.DataFrame(b2, columns=[ocean_cat[i] for i in 
range(len(ocean_cat))])

ocean_LabelBinarize

housing_prepared_new = pd.concat([new_features, ocean_LabelBinarize], 
axis=1)

print('Shape of new data prepared by above steps', 
housing_prepared_new.shape)

当我们使用管道时,它也会保留原始 (ocean_proximity) 变量和新创建的二值化列

【讨论】:

以上是关于sklearn 转换管道和功能联合的主要内容,如果未能解决你的问题,请参考以下文章

如何修复特征联合和管道中的元组对象错误(使用 sklearn 时)?

如何将 sklearn 管道转换为 pyspark 管道?

在 sklearn 管道中转换估计器的结果

mlflow 如何使用自定义转换器保存 sklearn 管道?

SKLEARN // 将 GridsearchCV 与列变换和管道相结合

如何将 sklearn 管道转换为普通代码?