pandas-type-detector 0.9.1

Automatic pandas DataFrame type detection with locale-aware parsing

Pandas Type Detector

A modular, extensible system for automatically detecting and converting pandas DataFrame column types using a strategy pattern with confidence scoring and locale-aware parsing.

Features

• Locale-aware parsing: Built-in support for PT-BR and EN-US numeric formats
• Modular architecture: Each detector handles a specific data type
• Confidence scoring: Get confidence levels for type detection decisions
• Error handling modes: Choose how to handle conversion errors (coerce, raise, ignore)
• Excel compatibility: Correctly handles data that Excel might misinterpret
• Extensible design: Easy to add new locales or data types
• Production ready: Successfully tested with 14,607+ rows in production

Installation

pip install pandas-type-detector

Quick Start

from pandas_type_detector import TypeDetectionPipeline
import pandas as pd

# Create sample DataFrame with PT-BR numeric data
data = {
    'receita': ['1.364,00', '343', '111,1', '1.950,00'],
    'nome': ['João', 'Maria', 'Pedro', 'Ana'],
    'ativo': ['sim', 'não', 'sim', 'não'],
    'data': ['2024-01-01', '2024-01-02', '2024-01-03', '2024-01-04']
}

df = pd.DataFrame(data)

# Initialize the pipeline with PT-BR locale
pipeline = TypeDetectionPipeline(locale="pt-br")

# Automatically detect and convert all column types
df_converted = pipeline.fix_dataframe_dtypes(df)

print(df_converted.dtypes)
# Output:
# receita      float64
# nome          string  
# ativo       boolean
# data     datetime64[ns]

Locale Support

PT-BR (Portuguese Brazil)

• Decimal separator: , (comma)
• Thousands separator: . (dot)
• Currency: R$, BRL
• Boolean values: sim/não, verdadeiro/falso
• Date formats: DD/MM/YYYY, YYYY-MM-DD

EN-US (English United States)

• Decimal separator: . (dot)
• Thousands separator: , (comma)
• Currency: $, USD
• Boolean values: yes/no, true/false
• Date formats: MM/DD/YYYY, YYYY-MM-DD

Advanced Usage

Error Handling Modes

# Coerce errors (default) - convert invalid values to NaN
pipeline = TypeDetectionPipeline(locale="pt-br", on_error="coerce")

# Raise exceptions on conversion errors
pipeline = TypeDetectionPipeline(locale="pt-br", on_error="raise")

# Ignore problematic columns - leave them unchanged
pipeline = TypeDetectionPipeline(locale="pt-br", on_error="ignore")

Individual Column Detection

# Get detailed information about type detection
result = pipeline.detect_column_type(df['receita'])

print(f"Detected type: {result.data_type.value}")
print(f"Confidence: {result.confidence:.2f}")
print(f"Metadata: {result.metadata}")

Skip Specific Columns

# Skip certain columns during conversion
df_converted = pipeline.fix_dataframe_dtypes(
    df, 
    skip_columns=['keep_as_string', 'manual_column']
)

Data Type Detection

The package detects the following data types:

Data Type	Description	Examples
INTEGER	Whole numbers	123, 1.000 (PT-BR)
FLOAT	Decimal numbers	123,45 (PT-BR), 123.45 (EN-US)
BOOLEAN	True/false values	sim/não, yes/no
DATETIME	Date and time	2024-01-01, 01/02/2024
TEXT	String data	Any non-matching text

Real-World Examples

Excel Import Fix

# Excel often misinterprets PT-BR numbers as dates
# This package correctly identifies and converts them

excel_data = ['1.364,00', '2.500,75', '3.100,25']
df = pd.DataFrame({'revenue': excel_data})

pipeline = TypeDetectionPipeline(locale="pt-br")
df_fixed = pipeline.fix_dataframe_dtypes(df)

# Now revenue is properly converted to float64
print(df_fixed['revenue'].tolist())
# Output: [1364.0, 2500.75, 3100.25]

Mixed Data Handling

# Handle mixed valid/invalid data gracefully
messy_data = ['1.364,00', 'invalid', '111,1', '-', '', '1.950,00']
df = pd.DataFrame({'values': messy_data})

pipeline = TypeDetectionPipeline(locale="pt-br", on_error="coerce")
df_clean = pipeline.fix_dataframe_dtypes(df)

# Invalid values become NaN, valid ones are converted
print(df_clean['values'].tolist())
# Output: [1364.0, NaN, 111.1, NaN, NaN, 1950.0]

Production ETL Pipeline

def process_financial_data(df):
    """Production ETL function using type detector."""
    
    # Configure for strict error handling in production
    pipeline = TypeDetectionPipeline(
        locale="pt-br", 
        on_error="raise",
        sample_size=1000
    )
    
    try:
        # Convert all columns automatically
        df_processed = pipeline.fix_dataframe_dtypes(df)
        
        # Log conversion results
        print(f"Successfully processed {len(df_processed)} rows")
        return df_processed
        
    except Exception as e:
        print(f"Data quality issue detected: {e}")
        raise

Architecture

The package uses a modular strategy pattern:

# Each detector handles one specific data type
from pandas_type_detector import (
    NumericDetector,    # Handles PT-BR/EN-US numbers
    BooleanDetector,    # Handles locale-specific booleans  
    DateTimeDetector,   # Handles various date formats
    TextDetector        # Fallback for text data
)

# All coordinated by the main pipeline
pipeline = TypeDetectionPipeline(locale="pt-br")

Configuration

Custom Sample Size

# Use larger sample for better accuracy on big datasets
pipeline = TypeDetectionPipeline(
    locale="pt-br",
    sample_size=5000  # Default: 1000
)

Confidence Thresholds

# Access individual detectors for custom configuration
from pandas_type_detector import NumericDetector, LOCALES

detector = NumericDetector(
    locale_config=LOCALES["pt-br"],
    min_confidence=0.9  # Higher threshold
)

Testing

Run the comprehensive test suite:

cd pandas-type-detector
python tests/test.py

The test suite includes:

• PT-BR numeric format validation
• Error handling verification
• Boolean detection tests
• DateTime parsing tests
• Excel compatibility tests
• Real-world scenario validation

Performance

• Optimized sampling: Only analyzes a configurable sample of rows
• Early exit: Stops detection when high confidence is reached
• Minimal overhead: Designed for production ETL pipelines
• Memory efficient: Processes columns independently

Contributing

The modular design makes it easy to contribute:

Adding a New Locale

from pandas_type_detector import LOCALES, LocaleConfig

# Add Spanish locale
LOCALES['es-es'] = LocaleConfig(
    name='es-es',
    decimal_separator=',',
    thousands_separator='.',
    currency_symbols=['€', 'EUR'],
    date_formats=[r'^\d{1,2}/\d{1,2}/\d{4}$']
)

Adding a New Detector

from pandas_type_detector import TypeDetector, DataType, DetectionResult

class URLDetector(TypeDetector):
    def detect(self, series):
        # Implementation here
        pass
    
    def convert(self, series):
        # Implementation here  
        pass

Requirements

• Python 3.7+
• pandas >= 1.0.0
• numpy >= 1.19.0

License

MIT License - see LICENSE file for details.

Acknowledgments

Developed to solve real-world data quality issues in Brazilian financial data processing. Successfully deployed in production handling 14,607+ rows of complex PT-BR formatted data.

Support

• Issues: GitHub Issues
• Documentation: This README
• Examples: See tests/test.py for comprehensive usage examples

---

Made with ❤️ for the pandas community