Sklearn SGDClassifier调用.fit()报ValueError问题咨询

问题描述 #

训练Sklearn的SGDClassifier时，将数组类型的name和age作为输入特征预测color，调用.fit()方法时出现"setting an array element with a sequence"错误，但当name和age为单个元素而非数组时无报错。需明确该错误含义，以及SGDClassifier是否支持嵌套数组作为输入。

复现代码 #

from sklearn.model_selection import train_test_split
import pandas as pd
import numpy as np

a=np.array([0, 2, 5, 2])
b=np.array( [0, 5, 0, 2])
c=np.array([2,2,0,0])
d=np.array([5,2,5,0])

age_a=np.array([5, 10, 7, 6])
age_b=np.array([3, 7, 11,8])
age_c=np.array([15, 10, 17, 2])
age_d=np.array([2, 8, 12,7])

data2={'name':[a,b,c,d],'age':[age_a, age_b, age_c, age_d],'color':[0,1,0,1]}
new2 = pd.DataFrame.from_dict(data2)

x = new2.loc[:, new2.columns != 'color']
y = new2.loc[:, 'color']

x=np.array(x,dtype=object)
y=np.array(y)

x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.25, random_state=42)

from sklearn.linear_model import SGDClassifier

sgd_clf=SGDClassifier(random_state=42)
sgd_clf.fit(x_train, y_train)
sgd_clf.predict(x_test)

报错信息 #

TypeError                                 Traceback (most recent call last)
TypeError: float() argument must be a string or a real number, not 'list'

The above exception was the direct cause of the following exception:

ValueError                                Traceback (most recent call last)
Cell In[117], line 25
     21 print(y_test)
     24 clf = SGDClassifier(loss="hinge", penalty="l2", max_iter=5)
     25 clf.fit(x_train, y_train)
     26 #SGDClassifier(max_iter=100)
     28 clf.predict([[2., 2.]])

ValueError: setting an array element with a sequence.

错误含义 #

这个错误的核心原因是：输入特征矩阵不符合Sklearn模型要求的标准二维数值数组格式。Sklearn期望输入的X是形状为[n_samples, n_features]的二维数组，其中每个元素都是单一的数值类型（如int、float）。而你的name和age列每个元素都是数组/序列，导致整个特征矩阵变成了嵌套的对象数组，模型无法将其转换为统一的数值矩阵，因此抛出错误。

SGDClassifier对输入的要求 #

SGDClassifier（以及所有Sklearn监督学习模型）不能直接接受嵌套数组作为输入特征。所有模型的输入特征必须是"扁平"的二维数值数组，每个样本对应一行，每个特征对应一列，且每个特征值为单个数值，不能是序列或数组。

解决方案 #

针对嵌套数组特征，有两种常见的处理方式：

1. 展开嵌套数组为多个特征 #

将每个嵌套数组中的元素拆分为单独的特征列，让特征矩阵变成标准的二维数值结构。示例代码如下：

from sklearn.model_selection import train_test_split
import pandas as pd
import numpy as np
from sklearn.linear_model import SGDClassifier

# 原始数据定义
a=np.array([0, 2, 5, 2])
b=np.array( [0, 5, 0, 2])
c=np.array([2,2,0,0])
d=np.array([5,2,5,0])

age_a=np.array([5, 10, 7, 6])
age_b=np.array([3, 7, 11,8])
age_c=np.array([15, 10, 17, 2])
age_d=np.array([2, 8, 12,7])

data2={'name':[a,b,c,d],'age':[age_a, age_b, age_c, age_d],'color':[0,1,0,1]}
new2 = pd.DataFrame.from_dict(data2)

# 展开name数组为4个特征列
new2['name_0'] = new2['name'].apply(lambda x: x[0])
new2['name_1'] = new2['name'].apply(lambda x: x[1])
new2['name_2'] = new2['name'].apply(lambda x: x[2])
new2['name_3'] = new2['name'].apply(lambda x: x[3])

# 展开age数组为4个特征列
new2['age_0'] = new2['age'].apply(lambda x: x[0])
new2['age_1'] = new2['age'].apply(lambda x: x[1])
new2['age_2'] = new2['age'].apply(lambda x: x[2])
new2['age_3'] = new2['age'].apply(lambda x: x[3])

# 选择展开后的特征，丢弃原始嵌套列
x = new2.drop(['color', 'name', 'age'], axis=1)
y = new2['color']

# 拆分数据集并训练
x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.25, random_state=42)
sgd_clf=SGDClassifier(random_state=42)
sgd_clf.fit(x_train, y_train)
print(sgd_clf.predict(x_test))

2. 提取嵌套数组的统计特征 #

如果不需要保留嵌套数组的每个元素，可以提取数组的统计量（如均值、最大值、最小值、方差等）作为单个特征。示例代码如下：

from sklearn.model_selection import train_test_split
import pandas as pd
import numpy as np
from sklearn.linear_model import SGDClassifier

# 原始数据定义
a=np.array([0, 2, 5, 2])
b=np.array( [0, 5, 0, 2])
c=np.array([2,2,0,0])
d=np.array([5,2,5,0])

age_a=np.array([5, 10, 7, 6])
age_b=np.array([3, 7, 11,8])
age_c=np.array([15, 10, 17, 2])
age_d=np.array([2, 8, 12,7])

data2={'name':[a,b,c,d],'age':[age_a, age_b, age_c, age_d],'color':[0,1,0,1]}
new2 = pd.DataFrame.from_dict(data2)

# 提取name数组的统计特征
new2['name_mean'] = new2['name'].apply(lambda x: np.mean(x))
new2['name_max'] = new2['name'].apply(lambda x: np.max(x))
new2['name_min'] = new2['name'].apply(lambda x: np.min(x))

# 提取age数组的统计特征
new2['age_mean'] = new2['age'].apply(lambda x: np.mean(x))
new2['age_std'] = new2['age'].apply(lambda x: np.std(x))
new2['age_median'] = new2['age'].apply(lambda x: np.median(x))

# 选择统计特征，丢弃原始嵌套列
x = new2.drop(['color', 'name', 'age'], axis=1)
y = new2['color']

# 拆分数据集并训练
x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.25, random_state=42)
sgd_clf=SGDClassifier(random_state=42)
sgd_clf.fit(x_train, y_train)
print(sgd_clf.predict(x_test))

内容的提问来源于stack exchange，提问作者Ash Ketchump