iterrows does not preserve dtypes

[George3d6]

Is your feature request related to a problem?

I wish I could use pandas in a way that does not cause bug-hunting for hours to look for a specific spot where nonsensical behavior occurs.

Describe the solution you'd like

The iterrow function (https://pandas.pydata.org/docs/reference/api/pandas.DataFrame.iterrows.html) and a few other ways of access do not preserve data types, this is very counterintuitive behavior (unless reading the docs, and making a comment about it for the next guy that maintain the code to see).

Fixing this would involve making sure they return the correct types and potentially adding a breaking_iterrow function that can be used for more efficient type-unaware iteration.

API breaking implications

Would involve changing the behavior of iterrows, might break a few pieces of code that rely upon the arbitrary types this generates (likely without the author realizing)

Describe alternatives you've considered

Leaving comments every time I use pandas iterrows.

[phofl]

Did you try itertuples? We specifically recommend it for this use case, see https://pandas.pydata.org/docs/reference/api/pandas.DataFrame.iterrows.html

[mroeschke]

Sorry for your frustration, but as stated in that docstring:

Because iterrows returns a Series for each row, it does not preserve dtypes across the rows

If you believe there's incorrect behavior, it would be useful for the team to investigate with minimum, copy-pastable example: https://matthewrocklin.com/blog/work/2018/02/28/minimal-bug-reports. A breaking_iterrow function would be out of scope.

[kwhkim]

It seems that iterrow is misleadning. How about iterseries or so. Naming is hard! I've been using iterrow but was not unaware of this problem.. :(

[phofl]

-1 on this

One column of a DataFrame is often viewed as a Series, this would be really confusing. The disadvantages of iterrows are clearly documented and an alternative is provided.

[kwhkim]

Sure. I understand what you mean. But One column of a DataFrame is naturally a Series, and one row of most DataFrames can not be a series without converting dtype. There could another options like iterseries(axis=0) or iterseries(axis=1).

[phofl]

If you do not want to change the dtype, just use itertuples

[jreback]

there are already too many iteration methods

-1 on any change

[kwhkim]

Thanks. but not quite the point though. I am talking about bad naming :(