pythonic way to get a zero-record slice of a pandas dataframe

Question

I have a pandas data frame, and I want to get a zero-record slice. That is, a dataframe with the same columns but zero rows. The reason I am doing this, is because i want to have an empty dataframe, to which i add rows from the original dataframe in a loop.

Currently if am using:

empty = df[0:0]

is this the pythonic way?

`pandas` provides tons of ways to avoid loops, are you sure you need one? — Lev Levitsky, Dec 27 '15 at 16:55

score 4 · Answer 1 · edited May 23 '17 at 11:59

4

Well, obvious way to make dataframe with known columns is to do

import pandas as pd
df = pd.DataFrame(columns=["A", "B", "C"])

You'll get empty dataframe as desired. But adding rows one by one is NOT most efficient way of operations

UPDATE

There was a discussion quite some time ago, take a look at add one row in a pandas.DataFrame

edited May 23 '17 at 11:59

Community

1
1

answered Dec 27 '15 at 17:04

Severin Pappadeux

18,636
3
38
64

so in my case you are suggesting: empty = pd.DataFrame(columns=df.columns) – o17t H1H' S'k Dec 27 '15 at 17:23
@eyaler if you really need it to do row-by-row, then yes – Severin Pappadeux Dec 27 '15 at 17:29
Note that this is different from a zero record slice, since it will have all object dtypes! – machow Apr 01 '22 at 03:29

score 1 · Answer 2 · answered Apr 01 '22 at 03:27

You can get a zero-record slice by indexing with something that returns no rows:

import pandas as pd

df = pd.DataFrame({"x": [1,2], "y": [1.2, 3.4]})

# select rows using an empty index, so get no rows back
res = df.loc[pd.Index([]), :]

Here's the result:

Empty DataFrame
Columns: [x]
Index: []

The accepted answer does not necessarily give back a zero-record slice of the original DataFrame. Its dtypes will all be object. This is not the case with the approach above!

We can check its dtypes to verify:

res.dtypes

Gives:

x      int64
y    float64
dtype: object

pythonic way to get a zero-record slice of a pandas dataframe

2 Answers2