sparse-profile 0.1.1

EDA on sparse data for classification problems

Sparse profile - EDA on sparse data

Module to perform EDA tasks for a classification problem with sparse data
Curently takes only numeric values

Sample usage

import pandas as pd
import numpy as np
from sparse_profile import sparse_profile

df = pd.DataFrame({
        'target' : [1, 1, 1, 1, 0, 0 ,0 ,0, 1, 0],
        'col_1' :  [1, 0, 0, 0, 0, 0, 0, 0, 0, 9],
        'col_2' :  [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
    })
sProfile = sparse_profile(df, 'target')
print(sProfile.top_gain)

Output maximum gain obtained from each column

col_2    0.422810
col_1    0.074882
dtype: float64

print(sProfile.report_sparsity)

Output percentage of zeros in column

col_1  0.8
col_2  0.1

Various sparse_profile reports can be accessed as attributes of the sparse_profile class object. List of all available attributes:

• report_sparsity:      pandas dataframe, Percentage of zeros in each column
• report_distinct:       pandas dataframe, Count of distinct non zero values in each column
• report_overall:        pandas dataframe, Overall summary of each column (similar to pandas describe())
• report_non_zero:    pandas dataframe, Summary of each column after removing zeros
• gain_df:                   pandas dataframe, Relative information gain at decile cutoffs for each column wrt target column
• auc_df:                    pandas dataframe, AUC of each column wrt target column
• top_gain:                pandas dataframe, Columns sorted by maximum gain obtained from gain_df