get_dummies(dummy_na=True) treats int like float when creating dummy names

[superxiao]

pd.get_dummies(pd.DataFrame({"id":[1,2,3]}), columns=["id"], dummy_na=True)
will produce

	id_1.0	id_2.0	id_3.0	id_nan
0	1	0	0	0
1	0	1	0	0
2	0	0	1	0

Which creates different column names from pd.get_dummies(pd.DataFrame({"id":[1,2,3]}), columns=["id"])

	id_1	id_2	id_3
0	1	0	0
1	0	1	0
2	0	0	1

It looks that this line

pandas/pandas/core/reshape/reshape.py

Line 909 in 6c0c277

levels = np.append(levels, np.nan)

converts the int array into float array because of np.nan.

[bobhaffner]

Hi @superxiao, it looks ok to me. Both lines produce dataframes with all dtypes being uint8

I'm running 0.22.0

Please paste the output of pd.show_verions() like instructed below

#### Output of ``pd.show_versions()``

<details>

[paste the output of ``pd.show_versions()`` here below this line]

</details>

[gfyoung]

@superxiao : I supposed we could always cast to a non-numeric dtype to prevent this.

@bobhaffner : This looks like regression to me then.

[gfyoung]

@bobhaffner : I get the same result as @superxiao on 0.22, so this doesn't look like a regression.

[bobhaffner]

@gfyoung Ok, yeah, I understand now that the title has changed. I was looking at the values themselves and not the column names

[psinger]

Encountering the same "issue", any idea how to prevent this?

[gfyoung]

Not sure ATM. The only workarounds I can think of:

1. Don't pass in dummy_na=True

2. Manually "convert" your columns after the fact.

🙁

[cjavier70]

Here's how I did the conversion....
dataframe_obj.rename(lambda col: col.replace('.0', ''), axis='columns')