Code Sample, a copy-pastable example if possible
index = np.linspace(-3, 5, 7)
df = pd.DataFrame({'A': np.sin(index), 'B': np.cos(index**2)}, index=index)
df.drop(columns=['A'])
Problem description
The docs say that I can use the columns
keyword, but when I try I get the following error:
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
<ipython-input-139-3a18b0426444> in <module>()
2 df = pd.DataFrame({'A': np.sin(index), 'B': np.cos(index**2)}, index=index)
3 # type(df.apply(np.sum, axis=1, reduce=False))
----> 4 df.drop(columns=['A'])
TypeError: drop() got an unexpected keyword argument 'columns'
The following is a workaround:
df.drop('A', axis=1)
Output of pd.show_versions()
Comment From: TomAugspurger
The index
/ columns
keywords were added in 0.21 (there's a little note about that just below those keywords in the parameters list).
Comment From: anuzis
Getting this error in pandas 0.24.0.
I have a working pandas script that runs fine on 0.23.4. I needed to make a new virtualenv with TensorFlow GPU support via conda and it automatically installed a newer version of pandas (0.24.0), however, the newer version isn't backward compatible using columns as a keyword argument for dropping in a dataframe.
Comment From: TomAugspurger
Can you make a new issue with a reproducible example?
On Thu, Jan 31, 2019 at 11:20 AM Michael Anuzis notifications@github.com wrote:
Getting this error in pandas 0.24.0.
I have a working pandas script that runs fine on 0.23.4. I needed to make a new virtualenv with TensorFlow GPU support via conda and it automatically installed a newer version of pandas (0.24.0), however, the newer version isn't backward compatible using columns as a keyword argument for dropping in a dataframe.
— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub https://github.com/pandas-dev/pandas/issues/19078#issuecomment-459429497, or mute the thread https://github.com/notifications/unsubscribe-auth/ABQHIp8Xr0Gw7HO5ti2-r9hKtbiGTxyzks5vIyX4gaJpZM4RTCbB .
Comment From: anuzis
Sure.
Edit: My mistake. Running 'pip freeze|grep pandas' from the conda virtualenv shows version 0.24.0, but running pd.show_versions() shows version 0.19.2... Haven't seen pip freeze show an incorrect mapping to a library version before...
'conda update pandas' shows it's 0.24.0, but somehow the conda virtualenv fails to use that version... I'll figure this out as it seems to be a non-pandas issue. Apologies for the mixup.
Edit2: If anyone else runs into the same issue the fix was running 'hash -r' from command line. ref: https://github.com/ipython/ipython/issues/10986
Comment From: DevinCharles
I'm running into this issue, on 0.24.2
INSTALLED VERSIONS
commit: None python: 3.7.2.final.0 python-bits: 64 OS: Windows OS-release: 10 machine: AMD64 processor: AMD64 Family 23 Model 1 Stepping 1, AuthenticAMD byteorder: little LC_ALL: None LANG: None LOCALE: None.None
pandas: 0.24.2
Comment From: DevinCharles
Hmmm... just realizing this is only happening on a multi-indexed df. Is there alternate syntax for that?
Comment From: TomAugspurger
@DevinCharles you can use a tuple
In [20]: df = pd.DataFrame({("A", 1): [1, 2], ("A", 2): [3, 4]})
In [21]: df.drop(columns=('A', 1))
Out[21]:
A
2
0 3
1 4
If you think there's still a bug, open a new issue, since this one has been closed.