将DataFrame列字符串值保留两位小数时，出现"无法将字符串'.'转换为浮点数"的ValueError报错该如何解决？

Problem Description #

I'm trying to round a string column in a pandas DataFrame to two decimal places. First I convert it to float using astype(float), then call round(2), with this code:

df['col'] = df['col'].astype(float).round(2)

But I get this error: ValueError: could not convert string to float: '.'. I thought decimal points wouldn't cause issues—what detail am I missing?

Additional notes:

• My dataset is large and may contain outliers, but the error still occurs even after filtering test samples.
• Later I found invalid data still exists even after filtering, and the solution provided by mozways works correctly.

Answer #

Ah, I’ve run into this exact headache before! The issue isn’t with normal decimal points in valid numbers (like "12.34")—it’s with isolated "." strings. Python’s float() function can’t parse a single dot because it has no numeric component before or after it; it’s not a valid floating-point value.

Here are two robust fixes tailored for this scenario:

1. Use pd.to_numeric with error coercion
   This is the go-to method for messy numeric string columns in pandas. It converts valid strings to floats automatically, and turns invalid entries (like ".", empty strings, or random text) into NaN, which you can handle (fill, drop, etc.) based on your needs:

   df['col'] = pd.to_numeric(df['col'], errors='coerce').round(2)
   

   If you want to replace NaN with a specific value (like 0) post-conversion, just add .fillna(0) at the end.

2. Preprocess to replace standalone dots
   If those single dots are supposed to represent 0.0 instead of NaN, you can target and replace them before conversion using regex:

   df['col'] = df['col'].replace('^\.$', '0.0', regex=True).astype(float).round(2)
   

   The regex ^\.$ matches only strings that are exactly a single dot, swapping them for "0.0" which converts cleanly to a float.

Since you mentioned mozways’ solution worked, it’s almost certainly one of these approaches—pd.to_numeric with errors='coerce' is built specifically to handle the edge cases that astype(float) can’t, making it perfect for large, messy datasets.

内容的提问来源于stack exchange，提问作者user19391566