Best Practice to ensure copy of a DataFrame (pandas)

Question

I have a pandas DataFrame, projthemes_df which contains three columns.

I want to subset it into a 2-column DataFrame. I've been using code like this because this is representative of the examples I see most often:

theme_by_code_df = projthemes_df[['code', 'name']]

This works.

There is duplication in the resulting DF.

When I tried

theme_by_code_df.drop_duplicates(inplace=True)

I got an error:

Apparently, the error is based on Returning a view versus a copy (although the link in the error message is incorrect).

The question:

I've been using

df2 = df1[['a', 'b', 'c']]

thinking I was getting a new DF in df2. OOps!

So, what's the best practice to ensure that I'm working with a DF I can safely modify?

I thought it would work to initialize an empty dataframe before doing the selection, but I got the same error with this code.

tmp = pd.DataFrame()
tmp = projthemes_df[['code', 'name']]
tmp.drop_duplicates(inplace=True)

Is this reasonable? Is there something simple/better?

tmp = pd.DataFrame(projthemes_df[['code', 'name']])

Check https://stackoverflow.com/questions/20625582/how-to-deal-with-settingwithcopywarning-in-pandas — BENY, Aug 04 '19 at 22:44
I use `copy`: `tmp = projthemes_df[['code', 'name']].copy()` — Trenton McKinney, Aug 04 '19 at 22:49

score 3 · Answer 1 · answered Aug 04 '19 at 23:32

3

Use the .copy() method. It will create a copy the data instead of giving you a slice of the original dataframe.

   tmp = projthemes_df[['code', 'name']].copy()

answered Aug 04 '19 at 23:32

sebvargo

1

Technically, it looks like (I hope) this will make a copy of a slice. – Vicki B Aug 06 '19 at 06:32
You are correct. In this example the copy is of the slice itself. – sebvargo Aug 06 '19 at 19:26

1 Answers1