Pandas-Profiling not considering the type of the column which is changed manually.

Describe the bug

• Pandas is used to read data
• One of the column was misread as number (int) instead of string/category (str)
• That column is manually typecast with the help of astype function [astype('str') / astype('category')]
• Pandas-Profiling is used to generate report
• The new type of the column is not considered and instead it considered that column as number.

To Reproduce GitHub repo: https://github.com/mohith7548/Pandas-Profiling-issue-recreation

import numpy as np
import pandas as pd
import pandas_profiling

df = pd.DataFrame({
    'Dummy': ['X', 'Y', 'X', 'X', 'Y', 'Y', 'Y', 'X'],
    'Contract': [9940243658, 9940243537, 9940243103, 9940242844, 9940242844, 9940242840, 9940242774, 9940242774]
})

df['Contract_category'] = df['Contract'].astype('category')
df['Contract_str'] = df['Contract'].astype('str')

df.dtypes
# output
Dummy                  object
Contract                int64
Contract_category    category
Contract_str           object
dtype: object

df.profile_report() # ProfileReport(df)
# When generating report, Pandas-Profiling has considered Contract_category & Contract_str as int 
# even though they are explicitly casted as category, str respectively above.

Version information: python version: 3.6 enviroment: jupyter-notebook pandas-profiling==2.10.0

Additional context As a work-around many suggested to prefix a character at the beginning of the number after type casting from int to str, so that Pandas-Profiling considers it as string/category. But this method seems to bring unnecessary complexity.

[ieaves]

@reedv the infer_dtypes flag attempts to infer the best data types for each column in your dataframe before computing summaries. For example, if you had a sequence of values [‘1’, ‘2’, ‘3’], when you use infer_dtypes, PP will provide a summary for integers rather than strings.

If you’ve already massaged your data and have things the way you want there’s no harm in calling it with infer_dtypes off. Under the hood, PP is using a customizable type system called visions and is moving towards fully customizable ProfileReports.

[mohith7548]

Thanks for the help @ieaves. I made a pull request making the developer choose infer/detect dtype by using a new kwarg param infer_dtypes in ProfileReport.
New changes here:
https://github.com/mohith7548/pandas-profiling/blob/4dd35dfbf7c1eb005a90e87275efdeec6968931b/src/pandas_profiling/model/summary.py#L46