to_numeric() for safe conversion in Pandas

Introduction

We use to_numeric() to change data into numbers safely. It helps avoid errors when some data can't be turned into numbers.

When you get data from a file and want to make sure columns are numbers.

When some values might be text but you want to convert what you can to numbers.

When you want to handle errors without stopping your program.

When cleaning messy data before analysis.

When you want to replace bad data with a default number.

Syntax

Pandas

pandas.to_numeric(arg, errors='raise', downcast=None)

arg is the data to convert (like a list or a DataFrame column).

errors controls what happens if conversion fails: 'raise' (stop), 'coerce' (make bad data NaN), or 'ignore' (keep original).

Examples

Converts a list of strings to numbers.

Pandas

pd.to_numeric(['1', '2', '3'])

Converts numbers and turns 'two' into NaN safely.

Pandas

pd.to_numeric(['1', 'two', '3'], errors='coerce')

Leaves the list as is because of 'ignore' error setting.

Pandas

pd.to_numeric(['1', 'two', '3'], errors='ignore')

Sample Program

This code shows how to convert a list of strings to numbers safely. The words 'thirty' and 'fifty' cannot be converted, so they become NaN (missing values) when using errors='coerce'.

Pandas

import pandas as pd

# Sample data with some non-numeric values
data = ['10', '20', 'thirty', '40', 'fifty']

# Convert with errors='raise' (default) - will cause error if uncommented
# pd.to_numeric(data)

# Convert with errors='coerce' to safely convert and replace bad data with NaN
converted = pd.to_numeric(data, errors='coerce')

print(converted)

Important Notes

Using errors='coerce' is helpful to avoid program crashes when data has unexpected text.

Result will be floats if there are any NaN values, because NaN is a float type.

You can use downcast to reduce memory, for example to 'integer' or 'float'.

Summary

to_numeric() safely converts data to numbers.

Use errors='coerce' to handle bad data by turning it into NaN.

This helps clean data before analysis without stopping your code.