Code Sample, a copy-pastable example if possible
As opposed to pandas, the default value for ddof
in standard deviation calculations is 0 in numpy. However, when I use groupby.agg(np.std)
I get the result for ddof=1
.
import numpy as np
import pandas as pd
prng = np.random.RandomState(0)
df = pd.DataFrame({"A": prng.choice(list("abcdef"), 15),
"B": prng.randint(0, 10, 15)})
df.groupby('A')['B'].agg(np.std)
Out[54]:
A
a 3.511885
b NaN
c 2.828427
d 3.593976
e 1.527525
f 1.414214
Name: B, dtype: float64
Expected Output
There is a note in the docs that states the default axis is changed but for me ddof change is quite unexpected. So my expected output is this:
df.groupby('A')['B'].apply(np.std)
Out[61]:
A
a 2.867442
b 0.000000
c 2.000000
d 3.112475
e 1.247219
f 1.000000
Name: B, dtype: float64
output of pd.show_versions()
INSTALLED VERSIONS
commit: None python: 3.5.1.final.0 python-bits: 64 OS: Windows OS-release: 10 machine: AMD64 processor: Intel64 Family 6 Model 60 Stepping 3, GenuineIntel byteorder: little LC_ALL: None LANG: en_GB
pandas: 0.18.1 nose: 1.3.7 pip: 8.1.1 setuptools: 20.3 Cython: 0.23.4 numpy: 1.11.0 scipy: 0.17.0 statsmodels: None xarray: None IPython: 4.1.2 sphinx: 1.3.1 patsy: 0.4.0 dateutil: 2.5.3 pytz: 2016.4 blosc: None bottleneck: 1.0.0 tables: 3.2.2 numexpr: 2.5 matplotlib: 1.5.1 openpyxl: 2.3.2 xlrd: 0.9.4 xlwt: 1.0.0 xlsxwriter: 0.8.4 lxml: 3.6.0 bs4: 4.4.1 html5lib: None httplib2: None apiclient: None sqlalchemy: 1.0.12 pymysql: None psycopg2: None jinja2: 2.8 boto: 2.39.0 pandas_datareader: None
Comment From: jreback
this is as expected (and has been these way for as long as I can remember).
numpy functions are translated to pandas ones. If you really want to use numpy then.
.groupby(...).agg(lambda x: np.std(x))
Comment From: jreback
you can also explicty pass in ddof=0
if you want.
Comment From: ayhanfuat
I thought about that possibility but seeing that it is not translated in df.groupby('A')['B'].apply(np.std)
I thought it could be a bug. Thanks.
Comment From: EricCousineau-TRI
you can also explicty pass in
ddof=0
if you want.
While some approaches use lambda x: np.std(x)
, you can also use functools.partial(np.std)
as a way to defeat func is np.std
(which is what seems to trigger this family of behaviors) - it also preserves the name.
FWIW Came here by way of StackOverflow post: https://stackoverflow.com/a/47699378/7829525
Comment From: stevetracvc
I'm posting here in case someone else runs into the same issue. I want both mean and standard deviation, but I can't pass in ddof to the std
function when using more than one aggregate function. So then, if I try to do something like:
df.agg(["mean", lambda x: np.std(x, ddof=0)])
That throws ValueError: cannot combine transform and aggregation operations
(same if I try to use partial
).
So instead I tried to do
df.agg([lambda x: np.mean(x), lambda x: np.std(x, ddof=0)])
But this actually does not work correctly because it's trying to transform each cell instead of doing an actual aggretate.
So this then turns into
df.agg(["mean", lambda x: np.std(x, axis=0, ddof=0)])