如何在Pandas DataFrame的convert_objects方法中设置例外项

Hey there! Let's tackle this problem properly—first off, note that convert_objects() is deprecated in newer pandas versions, so we'll use more reliable, up-to-date methods to get the dtype result you want. Here are two straightforward approaches:

Method 1: Convert only specified columns (Recommended) #

This approach is more efficient because it skips the columns you want to keep as object directly, avoiding unnecessary conversion and rollback steps:

import pandas as pd

# Define columns to exclude from numeric conversion
exclude_cols = ['device_id', 'email']

# Filter columns that need to be converted to numeric
convert_cols = [col for col in df.columns if col not in exclude_cols]

# Apply numeric conversion to target columns; unconvertible values become NaN
df[convert_cols] = df[convert_cols].apply(pd.to_numeric, errors='coerce')

# Check the resulting data types
print(df.dtypes)

After running this, you'll get exactly the dtype output you're looking for:

customer_id int64
device_id object
...
email object
email_counts float64
...
white_collar_count float64
dtype: object

Method 2: Full conversion first, then restore exception columns #

If you prefer a two-step process or already ran a full conversion, you can convert all columns first and then roll back the device_id and email columns to object type:

# Convert all columns to numeric (replaces deprecated convert_objects)
df = df.apply(pd.to_numeric, errors='coerce')

# Convert exception columns back to object type
df[['device_id', 'email']] = df[['device_id', 'email']].astype(object)

# Check the resulting data types
print(df.dtypes)

This will also produce your desired dtype result. Method 1 is generally better because it avoids risking unintended data changes (like non-numeric strings being turned into NaN and then becoming the string 'NaN' when converted back to object).

内容的提问来源于stack exchange，提问作者Nabih Bawazir