Shmessy

If your data is messy - Use Shmessy!

Shmessy designed to deal with messy pandas dataframes. We all knows the frustrating times when we as analysts or data-engineers should handle messy dataframe and analyze them by ourselves.

The goal of this tiny tool is to identify the physical / logical data type for each Dataframe column. It based on fast validators that will validate the data (Based on a sample) against regex / pydantic types or any additional validation function that you want to implement.

As you understand, this tool was designed to deal with dirty data, ideally developed for Dataframes generated from CSV / Flat files or any source that doesn't contain strict schema.

Installation

pip install shmessy

Usage

You have two ways to use this tool

Identify the Dataframe schema

import pandas as pd
from shmessy import Shmessy

df = pd.read_csv('/tmp/file.csv')
inferred_schema = Shmessy().infer_schema(df)

Output (inferred_schema dump):

{
    "infer_duration_ms": 12,
    "columns": [
        {
            "field_name": "id",
            "source_type": "Integer",
            "inferred_type": "Integer"
        },
        {
            "field_name": "email_value",
            "source_type": "String",
            "inferred_type": "Email"
        },
        {
            "field_name": "date_value",
            "source_type": "String",
            "inferred_type": "Date",
            "inferred_pattern": "%d-%m-%Y"
        },
        {
            "field_name": "datetime_value",
            "source_type": "String",
            "inferred_type": "Datetime",
            "inferred_pattern": "%Y/%m/%d %H:%M:%S"
        },
        {
            "field_name": "yes_no_data",
            "source_type": "String",
            "inferred_type": "Boolean",
            "inferred_pattern": [
                "YES",
                "NO"
            ]
        },
        {
            "field_name": "unix_value",
            "source_type": "Integer",
            "inferred_type": "UnixTimestamp",
            "inferred_pattern": "ms"
        },
        {
            "field_name": "ip_value",
            "source_type": "String",
            "inferred_type": "IPv4"
        }
    ]
}

Identify and fix Pandas Dataframe

This piece of code will change the column types of the input Dataframe according to Messy infer.

import pandas as pd
from shmessy import Shmessy

df = pd.read_csv('/tmp/file.csv')
fixed_df = Shmessy().fix_schema(df)

Original Dataframe

Fixed Dataframe

Read Messy CSV file

from shmessy import Shmessy
df = Shmessy().read_csv('/tmp/file.csv')

Original file

Fixed Dataframe

API

Constructor

shmessy = Shmessy(
    sample_size: Optional[int] = 1000,
    reader_encoding: Optional[str] = "UTF-8",
    locale_formatter: Optional[str] = "en_US",
    use_random_sample: Optional[bool] = True,
    types_to_ignore: Optional[List[str]] = None,
    max_columns_num: Optional[int] = 500,
    fallback_to_string: Optional[bool] = False,  # Fallback to string in case of casting exception
    fallback_to_null: Optional[bool] = False,  # Fallback to null in case of casting exception
    use_csv_sniffer: Optional[bool] = True,  # Use python sniffer to identify the dialect (seperator / quote-char / etc...)
    fix_column_names: Optional[bool] = False,  # Replace non-alphabetic/numeric chars with underscore
    numeric_types_max_length: Optional[int] = 20,  # Fallback to string for numeric values with many digits
)

read_csv

shmessy.read_csv(filepath_or_buffer: Union[str, TextIO, BinaryIO]) -> DataFrame

infer_schema

shmessy.infer_schema(df: Dataframe) -> ShmessySchema

fix_schema

shmessy.fix_schema(df: Dataframe) -> DataFrame

get_inferred_schema

shmessy.get_inferred_schema() -> ShmessySchema