将Pandas DataFrame转为rda文件时列数异常增多的问题求助

Hey there, I see you're hitting a frustrating issue where your 92-column Pandas DataFrame blows up to 1942 columns when saving to an RDA file using rpy2. Let's break down why this is happening and how to fix it.

What's Causing the Duplicate Columns? #

The problem likely stems from how the older pandas2ri interface handles conversions when globally activated. When you run pandas2ri.activate(), it sets up global conversion rules that can sometimes conflict—especially if there are unintended interactions between Pandas data types and R's data structures. In your test case, we can see each original column is duplicated multiple times (with .1, .2 suffixes), where each duplicate column is filled with a single value from the original column's rows. This is a clear sign the conversion logic is misinterpreting the DataFrame structure.

Solution 1: Use Local Conversion (Recommended for rpy2 3.x+) #

Instead of activating global conversions, use a local converter to handle the Pandas-to-R data frame conversion. This isolates the conversion process and avoids global conflicts. Here's how to adjust your code:

import pandas as pd
from rpy2 import robjects
from rpy2.robjects import pandas2ri
from rpy2.robjects.conversion import localconverter

def save_rdata_file(df, filename):
    # Use a local converter to avoid global conversion conflicts
    with localconverter(robjects.default_converter + pandas2ri.converter):
        r_data = robjects.conversion.py2rpy(df)
    # Assign the converted data frame to R environment and save
    robjects.r.assign('my_df', r_data)
    robjects.r(f"save(my_df, file='{filename}')")

# Test with your sample data
df = pd.DataFrame({'A': [1, 2, 3], 'B':[10, 9, 8], 'C':['A', 'B', 'C']})
save_rdata_file(df, 'test.rda')

This approach ensures the conversion is handled correctly without unintended side effects from global activation.

Solution 2: Use pyreadr (Simpler Alternative) #

If you want to avoid dealing with rpy2's conversion quirks entirely, use the pyreadr library. It's a lightweight tool built for reading/writing RDA/RData files directly from Pandas, no R environment required (it uses C++ bindings to R's file format under the hood).

First install it:

pip install pyreadr

Then save your DataFrame:

import pandas as pd
import pyreadr

df = pd.DataFrame({'A': [1, 2, 3], 'B':[10, 9, 8], 'C':['A', 'B', 'C']})
# Save the DataFrame as 'my_df' in the RDA file
pyreadr.write_rdata('test.rda', df, df_name='my_df')

This method is more straightforward and less prone to conversion bugs compared to rpy2 for simple save operations.

Verify the Fix #

After using either method, you can check the RDA file by loading it in R:

load('test.rda')
print(my_df)

You should see the original 3-column DataFrame with no duplicates, just like your input.

内容的提问来源于stack exchange，提问作者Zakary Krumlinde