使用Imputer将CSV中'-'替换为列均值时遇类型转换错误求助

Let's break down why you're getting that error and fix it step by step:

The Root of the Problem #

Your code has two key issues that lead to the could not convert string to float: '-' error:

1. You extracted X from the original data before handling the 'T' and '-' values, so X still contains these string markers instead of numeric values.
2. The old Imputer class (deprecated in scikit-learn 0.20+) can't directly process string-based missing value markers like "-"—it expects numeric data to calculate means.

Solution 1: Use Pandas Directly (Simpler for Data Cleaning) #

This approach leverages pandas' built-in tools to handle missing values intuitively:

import pandas as pd
import numpy as np

# Load and clean initial data
data = pd.read_csv("austin_weather.csv")
data = data.drop(['Events', 'Date'], axis=1)

# Replace all special string markers with NaN (pandas' recognized missing value)
data = data.replace(['-', 'T'], np.nan)

# Convert all columns to numeric type (critical for mean calculations)
data = data.astype(float)

# Fill missing values with the mean of their respective columns
data_filled = data.fillna(data.mean())

# Extract your X and Y arrays if needed
X = data_filled.iloc[:, :-1].values
Y = data_filled.iloc[:, 18].values

Solution 2: Use Scikit-learn's SimpleImputer (For ML Pipelines) #

If you need to integrate this into a scikit-learn workflow, use the modern SimpleImputer (the replacement for the old Imputer):

import pandas as pd
import numpy as np
from sklearn.impute import SimpleImputer

# Load and clean initial data
data = pd.read_csv("austin_weather.csv")
data = data.drop(['Events', 'Date'], axis=1)

# Replace special markers with NaN and convert to numeric
data = data.replace(['-', 'T'], np.nan)
data = data.astype(float)

# Split into features and target
X = data.iloc[:, :-1].values
Y = data.iloc[:, 18].values

# Initialize imputer to fill NaN with column means
imputer = SimpleImputer(missing_values=np.nan, strategy='mean')
X_filled = imputer.fit_transform(X)

Key Notes #

• Always replace string-based missing markers with np.nan first—this is the universal missing value indicator for pandas and scikit-learn.
• Converting columns to numeric type is non-negotiable: you can't calculate a mean for string data!
• Avoid the old Imputer class entirely—it's no longer maintained and lacks the flexibility of SimpleImputer.

内容的提问来源于stack exchange，提问作者Nadeem Ahmed