refineryframe

The goal of the package is to simplify life for data scientists, that have to deal with imperfect raw data. The package suppose to detect and clean unexpected values, while doubling as safeguard in production code based on predifined conditions that arise from business assumptions or any other source. The package is well suited to be an initial preprocessing step in ml pipelines situated between data gathering and training/scoring steps.

Developed by Kyrylo Mordan (c) 2023

Installation

Install refineryframe via pip with

pip install refineryframe

Documentation

The documentation can be found here.

Feature List

• refineryframe.refiner.Refiner.add_index_to_duplicate_columns - adds an index to duplicate column names in a pandas DataFrame.
• refineryframe.refiner.Refiner.check_col_names_types - checks if a given dataframe has the same column names as keys in a given dictionary and those columns have the same types as items in the dictionary.
• refineryframe.refiner.Refiner.check_date_format - checks if the values in the datetime columns of the input dataframe have the expected 'YYYY-MM-DD' format.
• refineryframe.refiner.Refiner.check_date_range - checks if dates are in expected ranges.
• refineryframe.refiner.Refiner.check_duplicate_col_names - checks for duplicate column names in a pandas DataFrame.
• refineryframe.refiner.Refiner.check_duplicates - checks for duplicates in a pandas DataFrame.
• refineryframe.refiner.Refiner.check_inf_values - counts the inf values in each column of a pandas DataFrame.
• refineryframe.refiner.Refiner.check_missing_types - takes a DataFrame and a dictionary of missing types as input, and searches for any instances of these missing types in each column of the DataFrame.
• refineryframe.refiner.Refiner.check_missing_values - counts the number of NaN, None, and NaT values in each column of a pandas DataFrame.
• refineryframe.refiner.Refiner.check_numeric_range - checks if numeric values are in expected ranges.
• refineryframe.refiner.Refiner.detect_unexpected_values - detects unexpected values in a pandas DataFrame.
• refineryframe.refiner.Refiner.get_refiner_settings - extracts values of parameters from refiner and saves them in dictionary for later use.
• refineryframe.refiner.Refiner.get_type_dict_from_dataframe - returns a dictionary or string representation of a dictionary containing the data types of each column in the given pandas DataFrame.
• refineryframe.refiner.Refiner.get_unexpected_exceptions_scaned - returns unexpected_exceptions with appropriate settings to the values in the dataframe.
• refineryframe.refiner.Refiner.replace_unexpected_values - replaces unexpected values in a pandas DataFrame with missing types.
• refineryframe.refiner.Refiner.set_refiner_settings - updates input parameters with values from provided settings dict.
• refineryframe.refiner.Refiner.set_type_dict - changes the data types of the columns in the given DataFrame based on a dictionary of intended data types.
• refineryframe.refiner.Refiner.set_types - changes the data types of the columns in the given DataFrame based on a dictionary of intended data types.
• refineryframe.refiner.Refiner.set_updated_dataframe - updates dataframe inside Refiner class. Usefull when some manipulations with the dataframe are done in between steps.

Simple package usage examples

Content:

• Initializing Refiner class
  • defining general conditions
• Use of simple general conditions
  • detecting column types
  • using independant conditions
  • detecting unexpected values
  • replacing unexpected values
  • moulding types
• Use of complex targeted conditions
  • to detect unexpected
  • to replace unexpected
• Refiner class settings
  • extracting settigns
  • scanning unexpected conditions
  • recreating identical class with settings

Creating example data (exceptionally messy dataframe)

import os 
import sys 
import numpy as np
import pandas as pd
import logging
sys.path.append(os.path.dirname(sys.path[0])) 
from refineryframe.refiner import Refiner
from refineryframe.demo import tiny_example

df = tiny_example['dataframe']
df

.dataframe tbody tr th {
    vertical-align: top;
}

.dataframe thead th {
    text-align: right;
}

Defining specification for the dataframe

MISSING_TYPES = tiny_example['MISSING_TYPES']
MISSING_TYPES

{'date_not_delivered': '1850-01-09',
 'date_other_missing_type': '1850-01-08',
 'numeric_not_delivered': -999,
 'character_not_delivered': 'missing'}

replace_dict = tiny_example['replace_dict']
replace_dict

{-996: -999, '1000-01-09': '1850-01-09'}

unexpected_exceptions = {
    "col_names_types": "NONE",
    "missing_values": ["NumericColumn_exepted"],
    "missing_types": "NONE",
    "inf_values": "NONE",
    "date_format": "NONE",
    "duplicates": "ALL",
    "date_range": "NONE",
    "numeric_range": "NONE"
}

Initializing Refiner class

tns = Refiner(dataframe = df,
              replace_dict = replace_dict,
              loggerLvl = logging.DEBUG,
              unexpected_exceptions_duv = unexpected_exceptions)

function for detecting column types

tns.get_type_dict_from_dataframe()

{'num_id': 'int64',
 'NumericColumn': 'float64',
 'NumericColumn_exepted': 'float64',
 'NumericColumn2': 'float64',
 'NumericColumn3': 'int64',
 'DateColumn': 'datetime64[ns]',
 'DateColumn2': 'datetime64[ns]',
 'DateColumn3': 'object',
 'CharColumn': 'object'}

adding expected types

types_dict_str = {'num_id' : 'int64', 
                   'NumericColumn' : 'float64', 
                   'NumericColumn_exepted' : 'float64', 
                   'NumericColumn2' : 'float64', 
                   'NumericColumn3' : 'int64', 
                   'DateColumn' : 'datetime64[ns]', 
                   'DateColumn2' : 'datetime64[ns]', 
                   'DateColumn3' : 'datetime64[ns]', 
                   'CharColumn' : 'object'}

Use of simple general conditions

Check independent conditions

tns.check_missing_types()
tns.check_missing_values()
tns.check_inf_values()
tns.check_col_names_types()
tns.check_date_format()
tns.check_duplicates()
tns.check_numeric_range()

WARNING:Refiner:Column DateColumn3: (1850-01-09) : 1 : 20.00%
WARNING:Refiner:Character score was lower then expected: 97.14 < 100
WARNING:Refiner:Column NumericColumn: (NA) : 2 : 40.00%
WARNING:Refiner:Column NumericColumn_exepted: (NA) : 2 : 40.00%
WARNING:Refiner:Column NumericColumn2: (NA) : 4 : 80.00%
WARNING:Refiner:Column DateColumn2: (NA) : 4 : 80.00%
WARNING:Refiner:Column CharColumn: (NA) : 2 : 40.00%
WARNING:Refiner:Missing values score was lower then expected: 53.33 < 100
WARNING:Refiner:Column NumericColumn: (INF) : 2 : 40.00%
WARNING:Refiner:Column NumericColumn_exepted: (INF) : 1 : 20.00%
WARNING:Refiner:Inf score was lower then expected: 94.0 < 100
WARNING:Refiner:Column DateColumn2 has non-date values or unexpected format.
WARNING:Refiner:Date format score was lower then expected: 50.0 < 100

moulding types

tns.set_types(type_dict = types_dict_str)

tns.get_type_dict_from_dataframe() 

{'num_id': 'int64',
 'NumericColumn': 'float64',
 'NumericColumn_exepted': 'float64',
 'NumericColumn2': 'float64',
 'NumericColumn3': 'int64',
 'DateColumn': 'datetime64[ns]',
 'DateColumn2': 'datetime64[ns]',
 'DateColumn3': 'datetime64[ns]',
 'CharColumn': 'object'}

Using the main function to detect unexpected values

tns.detect_unexpected_values(earliest_date = "1920-01-01",
                         latest_date = "DateColumn3")

DEBUG:Refiner:=== checking for column name duplicates
DEBUG:Refiner:=== checking column names and types
DEBUG:Refiner:=== checking for presence of missing values
WARNING:Refiner:Column CharColumn: (NA) : 2 : 40.00%
WARNING:Refiner:Column DateColumn2: (NA) : 4 : 80.00%
WARNING:Refiner:Column NumericColumn: (NA) : 2 : 40.00%
WARNING:Refiner:Column NumericColumn2: (NA) : 4 : 80.00%
WARNING:Refiner:Missing values score was lower then expected: 52.0 < 100
DEBUG:Refiner:=== checking for presence of missing types
WARNING:Refiner:Column DateColumn3: (1850-01-09) : 2 : 40.00%
WARNING:Refiner:Column NumericColumn_exepted: (-999) : 1 : 20.00%
WARNING:Refiner:Numeric score was lower then expected: 98.75 < 100
WARNING:Refiner:Date score was lower then expected: 96.0 < 100
DEBUG:Refiner:=== checking propper date format
WARNING:Refiner:Column DateColumn2 has non-date values or unexpected format.
WARNING:Refiner:Date format score was lower then expected: 66.67 < 100
DEBUG:Refiner:=== checking expected date range
WARNING:Refiner:** Not all dates in DateColumn are later than DateColumn3
WARNING:Refiner:Column DateColumn : future date : 4 : 80.00%
WARNING:Refiner:Future dates score was lower then expected: 80.0 < 100
DEBUG:Refiner:=== checking for presense of inf values in numeric colums
WARNING:Refiner:Column NumericColumn: (INF) : 2 : 40.00%
WARNING:Refiner:Column NumericColumn_exepted: (INF) : 1 : 20.00%
WARNING:Refiner:Inf score was lower then expected: 88.0 < 100
DEBUG:Refiner:=== checking expected numeric range
WARNING:Refiner:Percentage of passed tests: 53.85%

tns.duv_score

0.5384615384615384

Using function to replace unexpected values with missing types

tns.replace_unexpected_values(numeric_lower_bound = "NumericColumn3",
                                numeric_upper_bound = 4,
                                earliest_date = "1920-01-02",
                                latest_date = "DateColumn2",
                                unexpected_exceptions = {"irregular_values": "NONE",
                                                            "date_range": "DateColumn",
                                                            "numeric_range": "NONE",
                                                            "capitalization": "NONE",
                                                            "unicode_character": "NONE"})

DEBUG:Refiner:=== replacing missing values in category cols with missing types
DEBUG:Refiner:=== replacing all upper case characters with lower case
DEBUG:Refiner:=== replacing character unicode to latin
DEBUG:Refiner:=== replacing missing values in date cols with missing types
DEBUG:Refiner:=== replacing missing values in numeric cols with missing types
DEBUG:Refiner:=== replacing values outside of expected date range
DEBUG:Refiner:=== replacing values outside of expected numeric range
DEBUG:Refiner:** Usable values in the dataframe:  44.44%
DEBUG:Refiner:** Uncorrected data quality score:  32.22%
DEBUG:Refiner:** Corrected data quality score:  52.57%

Use of complex targeted conditions

unexpected_conditions = {
    '1': {
        'description': 'Replace numeric missing with with zero',
        'group': 'regex_columns',
        'features': r'^Numeric',
        'query': "{col} < 0",
        'warning': True,
        'set': 0
    },
    '2': {
        'description': "Clean text column from '-ing' endings and 'not ' beginings",
        'group': 'regex clean',
        'features': ['CharColumn'],
        'query': [r'ing', r'^not.'],
        'warning': False,
        'set': ''
    },
    '3': {
        'description': "Detect/Replace numeric values in certain column with zeros if > 2",
        'group': 'multicol mapping',
        'features': ['NumericColumn3'],
        'query': '{col} > 2',
        'warning': True,
        'set': 0
    },
    '4': {
        'description': "Replace strings with values if some part of the string is detected",
        'group': 'string check',
        'features': ['CharColumn'],
        'query': f"CharColumn.str.contains('cted', regex = True)",
        'warning': False,
        'set': 'miss'
    }
    }

- to detect unexpected values

tns.detect_unexpected_values(unexpected_conditions = unexpected_conditions)

DEBUG:Refiner:=== checking for column name duplicates
DEBUG:Refiner:=== checking column names and types
WARNING:Refiner:Incorrect data types:
WARNING:Refiner:Column num_id: actual dtype is object, expected dtype is int64
WARNING:Refiner:Dtypes score was lower then expected: 88.89 < 100
DEBUG:Refiner:=== checking for presence of missing values
DEBUG:Refiner:=== checking for presence of missing types
WARNING:Refiner:Column CharColumn: (missing) : 3 : 60.00%
WARNING:Refiner:Column DateColumn2: (1850-01-09) : 4 : 80.00%
WARNING:Refiner:Column DateColumn3: (1850-01-09) : 4 : 80.00%
WARNING:Refiner:Column NumericColumn: (-999) : 4 : 80.00%
WARNING:Refiner:Column NumericColumn_exepted: (-999) : 4 : 80.00%
WARNING:Refiner:Column NumericColumn2: (-999) : 5 : 100.00%
WARNING:Refiner:Column NumericColumn3: (-999) : 1 : 20.00%
WARNING:Refiner:Numeric score was lower then expected: 78.46 < 100
WARNING:Refiner:Date score was lower then expected: 84.0 < 100
WARNING:Refiner:Character score was lower then expected: 91.43 < 100
DEBUG:Refiner:=== checking propper date format
DEBUG:Refiner:=== checking expected date range
DEBUG:Refiner:=== checking for presense of inf values in numeric colums
DEBUG:Refiner:=== checking expected numeric range
DEBUG:Refiner:=== checking additional cons
DEBUG:Refiner:Replace numeric missing with with zero
WARNING:Refiner:Replace numeric missing with with zero :: 1
DEBUG:Refiner:Detect/Replace numeric values in certain column with zeros if > 2
WARNING:Refiner:Detect/Replace numeric values in certain column with zeros if > 2 :: 2
WARNING:Refiner:Percentage of passed tests: 69.23%

- to replace unexpected values

tns.replace_unexpected_values(unexpected_conditions = unexpected_conditions)

DEBUG:Refiner:=== replacing missing values in category cols with missing types
DEBUG:Refiner:=== replacing all upper case characters with lower case
DEBUG:Refiner:=== replacing character unicode to latin
DEBUG:Refiner:=== replacing with additional cons
DEBUG:Refiner:Replace numeric missing with with zero
DEBUG:Refiner:Clean text column from '-ing' endings and 'not ' beginings
DEBUG:Refiner:Detect/Replace numeric values in certain column with zeros if > 2
DEBUG:Refiner:Replace strings with values if some part of the string is detected
DEBUG:Refiner:=== replacing missing values in date cols with missing types
DEBUG:Refiner:=== replacing missing values in numeric cols with missing types
DEBUG:Refiner:=== replacing values outside of expected date range
DEBUG:Refiner:=== replacing values outside of expected numeric range
DEBUG:Refiner:** Usable values in the dataframe:  82.22%
DEBUG:Refiner:** Uncorrected data quality score:  88.89%
DEBUG:Refiner:** Corrected data quality score:  97.53%

tns.dataframe

.dataframe tbody tr th {
    vertical-align: top;
}

.dataframe thead th {
    text-align: right;
}

tns.detect_unexpected_values(unexpected_exceptions = {
    "col_names_types": "NONE",
    "missing_values": "NONE",
    "missing_types": "ALL",
    "inf_values": "NONE",
    "date_format": "NONE",
    "duplicates": "ALL",
    "date_range": "NONE",
    "numeric_range": "NONE"
})

DEBUG:Refiner:=== checking for column name duplicates
DEBUG:Refiner:=== checking column names and types
WARNING:Refiner:Incorrect data types:
WARNING:Refiner:Column num_id: actual dtype is object, expected dtype is int64
WARNING:Refiner:Dtypes score was lower then expected: 88.89 < 100
DEBUG:Refiner:=== checking for presence of missing values
DEBUG:Refiner:=== checking propper date format
DEBUG:Refiner:=== checking expected date range
DEBUG:Refiner:=== checking for presense of inf values in numeric colums
DEBUG:Refiner:=== checking expected numeric range
WARNING:Refiner:Percentage of passed tests: 90.00%

Scores

print(f'duv_score: {tns.duv_score :.4}')
print(f'ruv_score0: {tns.ruv_score0 :.4}')
print(f'ruv_score1: {tns.ruv_score1 :.4}')
print(f'ruv_score2: {tns.ruv_score2 :.4}')

duv_score: 0.9
ruv_score0: 0.8222
ruv_score1: 0.8889
ruv_score2: 0.9753

Refiner class settings

import os 
import sys 
import numpy as np
import pandas as pd
import logging
sys.path.append(os.path.dirname(sys.path[0])) 
from refineryframe.refiner import Refiner
from refineryframe.demo import tiny_example

Initializing Refiner class

tns = Refiner(dataframe = tiny_example['dataframe'],
              replace_dict = tiny_example['replace_dict'],
              loggerLvl = logging.DEBUG,
              unexpected_exceptions_duv = {
                                            "col_names_types": "NONE",
                                            "missing_values": "ALL",
                                            "missing_types": "ALL",
                                            "inf_values": "NONE",
                                            "date_format": "NONE",
                                            "duplicates": "ALL",
                                            "date_range": "NONE",
                                            "numeric_range": "ALL"
                                        })

using the main function to detect unexpected values

tns.detect_unexpected_values()

DEBUG:Refiner:=== checking for column name duplicates
DEBUG:Refiner:=== checking column names and types
DEBUG:Refiner:=== checking propper date format
WARNING:Refiner:Column DateColumn2 has non-date values or unexpected format.
WARNING:Refiner:Date format score was lower then expected: 50.0 < 100
DEBUG:Refiner:=== checking expected date range
DEBUG:Refiner:=== checking for presense of inf values in numeric colums
WARNING:Refiner:Column NumericColumn: (INF) : 2 : 40.00%
WARNING:Refiner:Column NumericColumn_exepted: (INF) : 1 : 20.00%
WARNING:Refiner:Inf score was lower then expected: 88.0 < 100
WARNING:Refiner:Percentage of passed tests: 71.43%

extracting Refiner settings

refiner_settings = tns.get_refiner_settings()
refiner_settings

{'replace_dict': {-996: -999, '1000-01-09': '1850-01-09'},
 'MISSING_TYPES': {'date_not_delivered': '1850-01-09',
  'numeric_not_delivered': -999,
  'character_not_delivered': 'missing'},
 'expected_date_format': '%Y-%m-%d',
 'mess': 'INITIAL PREPROCESSING',
 'shout_type': 'HEAD2',
 'logger_name': 'Refiner',
 'loggerLvl': 10,
 'dotline_length': 50,
 'lower_bound': -inf,
 'upper_bound': inf,
 'earliest_date': '1900-08-25',
 'latest_date': '2100-01-01',
 'ids_for_dedup': 'ALL',
 'unexpected_exceptions_duv': {'col_names_types': 'NONE',
  'missing_values': 'ALL',
  'missing_types': 'ALL',
  'inf_values': 'NONE',
  'date_format': 'NONE',
  'duplicates': 'ALL',
  'date_range': 'NONE',
  'numeric_range': 'ALL'},
 'unexpected_exceptions_ruv': {'irregular_values': 'NONE',
  'date_range': 'NONE',
  'numeric_range': 'NONE',
  'capitalization': 'NONE',
  'unicode_character': 'NONE'},
 'unexpected_exceptions_error': {'col_name_duplicates': False,
  'col_names_types': False,
  'missing_values': False,
  'missing_types': False,
  'inf_values': False,
  'date_format': False,
  'duplicates': False,
  'date_range': False,
  'numeric_range': False},
 'thresholds': {'cmt_scores': {'numeric_score': 100,
   'date_score': 100,
   'cat_score': 100},
  'cmv_scores': {'missing_values_score': 100},
  'ccnt_scores': {'missing_score': 100, 'incorrect_dtypes_score': 100},
  'inf_scores': {'inf_score': 100},
  'cdf_scores': {'date_format_score': 100},
  'dup_scores': {'row_dup_score': 100, 'key_dup_score': 100},
  'cnr_scores': {'low_numeric_score': 100, 'upper_numeric_score': 100},
  'cdr_scores': {'early_dates_score': 100, 'future_dates_score': 100}},
 'unexpected_conditions': None,
 'ignore_values': [],
 'ignore_dates': [],
 'type_dict': {}}

Initializing new clean Refiner

tns2 = Refiner(dataframe = tiny_example['dataframe'])

scanning dataframe for unexpected conditions

scanned_unexpected_exceptions = tns2.get_unexpected_exceptions_scaned()
scanned_unexpected_exceptions

WARNING:Refiner:Column CharColumn: (NA) : 2 : 40.00%
WARNING:Refiner:Column DateColumn2: (NA) : 4 : 80.00%
WARNING:Refiner:Column NumericColumn: (NA) : 2 : 40.00%
WARNING:Refiner:Column NumericColumn_exepted: (NA) : 2 : 40.00%
WARNING:Refiner:Column NumericColumn2: (NA) : 4 : 80.00%
WARNING:Refiner:Missing values score was lower then expected: 53.33 < 100
WARNING:Refiner:Column DateColumn3: (1850-01-09) : 1 : 20.00%
WARNING:Refiner:Character score was lower then expected: 97.14 < 100
WARNING:Refiner:Column DateColumn2 has non-date values or unexpected format.
WARNING:Refiner:Date format score was lower then expected: 50.0 < 100
WARNING:Refiner:Column NumericColumn: (INF) : 2 : 40.00%
WARNING:Refiner:Column NumericColumn_exepted: (INF) : 1 : 20.00%
WARNING:Refiner:Inf score was lower then expected: 88.0 < 100
WARNING:Refiner:Percentage of passed tests: 73.33%





{'col_names_types': 'NONE',
 'missing_values': 'ALL',
 'missing_types': 'ALL',
 'inf_values': 'ALL',
 'date_format': 'ALL',
 'duplicates': 'NONE',
 'date_range': 'NONE',
 'numeric_range': 'NONE'}

detection before applying settings

tns2.detect_unexpected_values()

WARNING:Refiner:Column CharColumn: (NA) : 2 : 40.00%
WARNING:Refiner:Column DateColumn2: (NA) : 4 : 80.00%
WARNING:Refiner:Column NumericColumn: (NA) : 2 : 40.00%
WARNING:Refiner:Column NumericColumn_exepted: (NA) : 2 : 40.00%
WARNING:Refiner:Column NumericColumn2: (NA) : 4 : 80.00%
WARNING:Refiner:Missing values score was lower then expected: 53.33 < 100
WARNING:Refiner:Column DateColumn3: (1850-01-09) : 1 : 20.00%
WARNING:Refiner:Character score was lower then expected: 97.14 < 100
WARNING:Refiner:Column DateColumn2 has non-date values or unexpected format.
WARNING:Refiner:Date format score was lower then expected: 50.0 < 100
WARNING:Refiner:Column NumericColumn: (INF) : 2 : 40.00%
WARNING:Refiner:Column NumericColumn_exepted: (INF) : 1 : 20.00%
WARNING:Refiner:Inf score was lower then expected: 88.0 < 100
WARNING:Refiner:Percentage of passed tests: 73.33%

using saved refiner settings for new instance

tns2.set_refiner_settings(refiner_settings)

tns2.detect_unexpected_values()

DEBUG:Refiner:=== checking for column name duplicates
DEBUG:Refiner:=== checking column names and types
DEBUG:Refiner:=== checking propper date format
WARNING:Refiner:Column DateColumn2 has non-date values or unexpected format.
WARNING:Refiner:Date format score was lower then expected: 50.0 < 100
DEBUG:Refiner:=== checking expected date range
DEBUG:Refiner:=== checking for presense of inf values in numeric colums
WARNING:Refiner:Column NumericColumn: (INF) : 2 : 40.00%
WARNING:Refiner:Column NumericColumn_exepted: (INF) : 1 : 20.00%
WARNING:Refiner:Inf score was lower then expected: 88.0 < 100
WARNING:Refiner:Percentage of passed tests: 71.43%

tns3 = Refiner(dataframe = tiny_example['dataframe'], 
               unexpected_exceptions_duv = scanned_unexpected_exceptions)

tns3.detect_unexpected_values()
print(f'duv score: {tns3.duv_score}')

duv score: 1.0