Sweetviz Globally Alters NumPy Error Handling, Causing Underflow Errors

[smolenski9ld]

The sweetviz workaround at [graph_numeric.py] modifies global NumPy settings, causing potential fragility and failures in other Python modules (e.g., scikit-learn) that rely on NumPy.

[smolenski9ld]

Reproduction code:

import pandas as pd
import numpy as np
from sklearn.impute import SimpleImputer
import sweetviz as sv

def check_imputer(quiet = False):
    data = np.array([44.,  1.,  6.,  6.,  1.,  1., 72.,  1.,  4.,  1.], dtype=np.float32)
    column_name = 'fd_1'  # You can choose any name for your column
    
    df_X = pd.DataFrame({column_name: data})
    
    data_y = np.array([1, 1, 1, 1, 1, 1, 1, 1, 1, 1])
    series_y = pd.Series(data_y)
    
    new_simple_imputer = SimpleImputer()
    
    print(new_simple_imputer)
    
    new_simple_imputer.fit(X=df_X, y=series_y)
    print("Fit successful on the new imputer.")
    

dft2 = pd.DataFrame({"val": range(11)})
report = sv.analyze(dft2)
check_imputer(quiet = True)