check_int_size_requirements.py

import numpy as np
import pandas as pd

def check_var_size_requirements(df):
    ''' for integer columns, display smallest int type which could safely store values'''
    
    # Iterate through int columns 
    # (pandas' read_* is good about assigning some int type if all vals are ints and no NaNs)
    int_types = ['int', np.int8, np.int16, np.int32, np.int64, 
                 np.uint8, np.uint16, np.uint32, np.uint64]

    df = df.select_dtypes(include=int_types)
    
    for col_name in df.columns:
                
        # find max / min col values
        col_min = df[col_name].min()
        col_max = df[col_name].max()
        
        # decide smallest int datatype which can hold max/min values above
        if col_min >= 0:
            # (can use unsigned ints)
            use_type = ('uint8' if col_max < np.iinfo(np.uint8).max  else 
                       'uint16' if col_max < np.iinfo(np.uint16).max else 
                       'uint32' if col_max < np.iinfo(np.uint32).max else 
                       'uint64')
        else:
            # (have negative numbers, need signed integer types)
            use_type = ('int8'  if col_max < np.iinfo(np.int8).max  and col_min > np.iinfo(np.int8).min  else 
                        'int16' if col_max < np.iinfo(np.int16).max and col_min > np.iinfo(np.int16).min else 
                        'int32' if col_max < np.iinfo(np.int32).max and col_min > np.iinfo(np.int32).min else 
                        'int64')
                    
        # (For now I prefer this to be informational, instead of automatically recasting types.)
        print("{}\ncurrent type {}, could fit safely in type {} \n".format(col_name, 
                                                                           df[col_name].dtype, 
                                                                           use_type))


########################
# EXAMPLE / TEST
########################
#test_df = pd.DataFrame({'a': [100],
#                        'b': [1000],
#                        'c': [10000],
#                        'd': [100000],
#                        'e': [-150],
#                        'f': [100]})
#    
#test_df['f'] = test_df['f'].astype(np.int32)
#    
#check_var_size_requirements(test_df)

# OUTPUT:
#a
#current type int64, could fit safely in type uint8 
#
#b
#current type int64, could fit safely in type uint16 
#
#c
#current type int64, could fit safely in type uint16 
#
#d
#current type int64, could fit safely in type uint32 
#
#e
#current type int64, could fit safely in type int16 
#
#f
#current type int32, could fit safely in type uint8