TypeError: > not supported between instances of float and str in Scikit-Learn

Verified FixPython 3.10+Scikit-Learn 1.4+Silo: scikit-learn

Quick Fix / Solution Rapide

Clean mixed-type columns with .astype(str) or process numerical and categorical features separately using ColumnTransformer.

Root Cause Analysis

This error occurs when Python tries to sort or fit categorical transformer classes (like LabelEncoder or OrdinalEncoder) on a column containing mixed data types (such as both strings and floating-point NaN values).

1. Comparison Protocol in Transformers

When LabelEncoder.fit() is called, it sorts unique classes internally using Python's native < and > comparison operators to establish a deterministic integer mapping. In Python 3, comparing a str directly to a float (such as np.nan) is illegal and raises TypeError: '>' not supported between instances of 'float' and 'str'.

2. The Missing Value Trap in Text Columns

When pandas reads CSV files with empty cells in text columns, it fills missing entries with np.nan (a float) by default. The resulting column has object dtype containing both string labels ('Paris', 'London') and float values (NaN), which crashes LabelEncoder.

3. Resolving Mixed Types

Cast the entire column to string with df['col'].astype(str) or fill missing values before encoding with df['col'].fillna('Missing').

4. Using OrdinalEncoder for Features

Remember that LabelEncoder is intended exclusively for 1D target labels y. For feature matrices X, always use OrdinalEncoder or OneHotEncoder within a ColumnTransformer.

Reproduction Code (MCVE)

Example: Bug Reproduction

import numpy as np
from sklearn.preprocessing import LabelEncoder

# Mixed-type array containing both strings and floats (np.nan)
mixed_labels = np.array(['cat', 1.5, 'dog'], dtype=object)
le = LabelEncoder()

# TypeError: '<' not supported between instances of 'float' and 'str'
le.fit(mixed_labels)

Solution 1: Cast Array to Uniform String Dtype

Convert all elements in the column or array to homogeneous strings before fitting the encoder.

Example: Recommended Solution

import numpy as np
from sklearn.preprocessing import LabelEncoder

mixed_labels = np.array(['cat', 1.5, 'dog'], dtype=object)

# Convert all items to clean strings
clean_labels = mixed_labels.astype(str)

le = LabelEncoder()
encoded = le.fit_transform(clean_labels)
print('Encoded classes:', le.classes_)
print('Encoded output:', encoded)

Solution 2: Handle Missing Categories via SimpleImputer & OrdinalEncoder

Use SimpleImputer with a constant placeholder before categorical encoding in a structured Pipeline.

Example: Alternative Solution

import numpy as np
from sklearn.impute import SimpleImputer
from sklearn.preprocessing import OrdinalEncoder
from sklearn.pipeline import make_pipeline

X_raw = np.array([['red'], [np.nan], ['blue']], dtype=object)

pipeline = make_pipeline(
    SimpleImputer(strategy='constant', fill_value='missing'),
    OrdinalEncoder()
)
encoded = pipeline.fit_transform(X_raw)
print('Pipeline encoded output:', encoded)

A common mistake is using LabelEncoder on 2D feature matrices X, which causes shape errors and mixed-type sorting failures. LabelEncoder is strictly designed for 1D target vectors y. Use OneHotEncoder or OrdinalEncoder for features X. Contrast this TypeError with TypeError: '<' not supported between instances of 'str' and 'int'. Always check column types using df.dtypes and verify df.isna().sum() during preprocessing.

Previous Next