mealprep 1.0.11

Python package that ease the pain in pre-processing like outlier finding, numerical/categorical data and etc.

mealprep

Mealprep offers a toolkit, made with care, to help users save time in the data preprocessing kitchen.

Overview

Recognizing that the preparation step of a data science project often requires the most time and effort, mealprep aims to help data science chefs of all specialties master their recipes of analysis. This package tackles pesky tasks such as classifying columns as categorical or numeric ingredients, straining NA values and outliers, and automating a preprocessing recipe pipeline.

Functions

find_fruits_veg(): This function will drop rows with NAs and find the indices of columns with all numeric values or categorical values based on the specification.

find_missing_ingredients(): For each column with missing values, this function will create a reference list of row indices, sum the number, and calculate the proportion of missing values.

find_bad_apples(): This function uses a univariate approach to outlier detection. For each column with outliers (values that are 2 or more standard deviations from the mean), this function will create a reference list of row indices with outliers, and the total number of outliers in that column.

make_recipe(): This function is used to quickly apply common data preprocessing techniques.

Mealprep and Python’s Ecosystem

mealprep complements many of the existing packages in the Python ecosystem around the theme of data preprocessing. When preparing a dataframe for a machine learning preprocessing pipeline, it is time consuming to manually note which columns are categorical and numerical, particularly for large datasets. The pandas function df.select_dtypes() comes close by allowing users to select columns with data corresponding to specific data types however the output of this function is a pandas dataframe. find_fruits_veg() aims to fill this void by producing a list of columns corresponding to the categorical and numerical groups.

In terms of missing values, pandas package’s isna() function converts all elements of a pandas.dataframe or pandas.series to boolean values representing if they are missing values. The package autoimpute provides a suite of tools to fill missing values in a dataset through multiple univariate, multivariate and time series methods. The gap between these packages is that neither provides you a summary of the missing values including the list of indices where they occur. find_missing_ingredients() augments these tools by providing a summary dataframe detailing which columns have missing values, as well as their count and proportion.

The pandas package’s describe() function is a staple in the data wrangling process because it returns several summary statistics for each numeric column in a dataframe, such as the mean, standard deviation, minimum, and maximum. Viewing these statistics together is helpful for detecting outliers. However, the output of this function does not tell you which rows of data these outliers are found in, or how many outliers are present in the dataframe. Packages like the PyOD toolkit and other functions that use clustering methods consider all variables at once to detect outliers for multivariate data. PyOD provides over 20 algorithms to select from in detecting these outliers, which is handy for large multivariate datasets where you know you want to consider all features in detecting outliers, but can be a bit extreme for initial data exploration. The mealprep find_bad_apples() function lives happily in the space between pandas and PyOD-type solutions for outlier detection, where it provides more information than the pandas describe() function to point out datapoints which need further investigation, but does not consider all variables at once like the PyOD-type functions do.

Lastly, there are many great tools in the data science ecosystem for pre-processing data such as scikit-learn preprocessing in Python. However, you may find yourself frequently writing the same lengthy code for common preprocessing tasks (e.g scale numeric features and one hot encode categorical features). preprocess_recipe() provides a shortcut function to apply your favourite recipes quickly to preprocess data in one line of code.

Installation:

pip install mealprep

Examples

find_fruits_veg()

Find the column indices for either numerical or categorical variables in your dataframe with the find_fruits_veg() function. The example below shows how to use find_fruits_veg() to find the index of the categorical column in a toy dataframe.

First, load the required packages.

from mealprep.mealprep import find_fruits_veg
import pandas as pd

If you don’t already have a dataframe to work with, run this code to set up a toy dataframe (df) for testing.

df = pd.DataFrame({'col1': [1, 2], 'col2': ['a', 'b']})
df

##    col1 col2
## 0     1    a
## 1     2    b

Then, apply the find_fruits_veg() function to the dataframe.

find_fruits_veg(df, type_of_out = 'categ')

## [1]

find_missing_ingredients()

Before launching into a new data analysis, running the function find_missing_ingredients() on a dataframe of interest will produce a report on each column with missing values.

First, load the required packages

from mealprep.mealprep import find_missing_ingredients
import pandas as pd
import numpy as np

If you don’t already have a dataframe to work with, run this code to set up a toy dataframe (df) for testing.

test1= {'column1': ['a', 'b', 'c', 'd'],
       'column2': [1, 2, np.NaN, 3],
       'column3': [np.NaN] * 4}
   
df = pd.DataFrame(test1)
df

##   column1  column2  column3
## 0       a      1.0      NaN
## 1       b      2.0      NaN
## 2       c      NaN      NaN
## 3       d      3.0      NaN

Then, apply the find_missing_ingredients() function to the dataframe.

find_missing_ingredients(df)

##   Column name  NaN count NaN proportion   NaN indices
## 0     column2          1          25.0%           [2]
## 1     column3          4         100.0%  [0, 1, 2, 3]

find_bad_apples()

Find the outliers in your data by applying the find_bad_apples() function to your dataframe.

First, load the required packages.

from mealprep.mealprep import find_bad_apples
import pandas as pd

If you don’t already have a dataframe to work with, run this code to set up a toy dataframe (df) for testing.

df = pd.DataFrame({'A' : [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
                             1,1,1,1,1,1,1,1,1,1,1,1,1,1,1],
                    'B' : [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,-100,
                             1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,100],
                    'C' : [1,1,1,1,1,19,1,1,1,1,1,1,1,1,19,1,1,1,1,
                             1,1,1,1,1,1,1,19,1,1,1,1,1,1,1,1]})
df

##     A    B   C
## 0   1    1   1
## 1   1    1   1
## 2   1    1   1
## 3   1    1   1
## 4   1    1   1
## 5   1    1  19
## 6   1    1   1
## 7   1    1   1
## 8   1    1   1
## 9   1    1   1
## 10  1    1   1
## 11  1    1   1
## 12  1    1   1
## 13  1    1   1
## 14  1    1  19
## 15  1    1   1
## 16  1    1   1
## 17  1 -100   1
## 18  1    1   1
## 19  1    1   1
## 20  1    1   1
## 21  1    1   1
## 22  1    1   1
## 23  1    1   1
## 24  1    1   1
## 25  1    1   1
## 26  1    1  19
## 27  1    1   1
## 28  1    1   1
## 29  1    1   1
## 30  1    1   1
## 31  1    1   1
## 32  1    1   1
## 33  1    1   1
## 34  1  100   1

Then, apply the find_bad_apples() function to the dataframe.

find_bad_apples(df)

##   Variable      Indices Total Outliers
## 0        B     [17, 34]              2
## 1        C  [5, 14, 26]              3

make_recipe()

Do you find yourself constantly applying the same data preprocessing techniques time and time again? make_recipe can help by applying your favourite preprocessing recipes in only a few lines of code.

Below make_recipe applies the following common recipe in only one line of code:

1. Split data into training, validation, and testing
2. Standardise and scale numeric features
3. One hot encode categorical features

First, load the required packages.

from mealprep.mealprep import make_recipe
import pandas as pd
import numpy as np
from vega_datasets import data

If you don’t already have a dataframe to work with, run this code to load the classic mtcars dataset for testing.

df = pd.read_json(data.cars.url).drop(columns=["Year"])
X = df.drop(columns=["Name"])
y = df[["Name"]]
    
df.info()

## <class 'pandas.core.frame.DataFrame'>
## RangeIndex: 406 entries, 0 to 405
## Data columns (total 8 columns):
##  #   Column            Non-Null Count  Dtype  
## ---  ------            --------------  -----  
##  0   Name              406 non-null    object 
##  1   Miles_per_Gallon  398 non-null    float64
##  2   Cylinders         406 non-null    int64  
##  3   Displacement      406 non-null    float64
##  4   Horsepower        400 non-null    float64
##  5   Weight_in_lbs     406 non-null    int64  
##  6   Acceleration      406 non-null    float64
##  7   Origin            406 non-null    object 
## dtypes: float64(4), int64(2), object(2)
## memory usage: 25.5+ KB

Then, use make_recipe to quickly apply split your data and apply your favourite preprocessing techniques!

X_train, X_valid, X_test, y_train, y_valid, y_test = make_recipe(
    X=X, y=y, recipe="ohe_and_standard_scaler", 
    splits_to_return="train_test")

X_train.head()

##    Miles_per_Gallon  Cylinders  Displacement  ...  x0_Europe  x0_Japan  x0_USA
## 0          1.136396  -0.825761     -0.900964  ...        0.0       0.0     1.0
## 1         -1.084286   1.441582      1.652002  ...        0.0       0.0     1.0
## 2         -0.305099  -0.825761     -0.691131  ...        0.0       0.0     1.0
## 3         -1.344015   1.441582      1.311023  ...        0.0       0.0     1.0
## 4         -0.564828  -0.825761     -0.691131  ...        0.0       0.0     1.0
## 
## [5 rows x 9 columns]

Documentation

The official documentation is hosted on Read the Docs: https://mealprep.readthedocs.io/en/latest/

Credits

This package was created with Cookiecutter and the UBC-MDS/cookiecutter-ubc-mds project template, modified from the pyOpenSci/cookiecutter-pyopensci project template and the audreyr/cookiecutter-pypackage.