Why does pandas Round method explodes my data frame?

[blucap]

I try to round all values in this dataframe. However, the pandas round() method explodes my dataframe from 150 rows to 7518 rows.

Perhaps there is something odd with the data in the dataframe, but then again, one would not expect a simple rounding function to do this.

Below, I replicate the error using 1) simulated data and 2) the data that leads to the said error.

This results in 150 rows, which is the correct number:

import pandas as pd
import numpy as np
df = pd.DataFrame(np.random.random([150, 4]), columns=['A', 'B', 'C', 'D'])
df["cat"] = "MID"
df.loc[:399,["cat"]] = "LOW"
df.iloc[-400:,-1] = "HI"
df.cat.value_counts()
df.set_index("cat", inplace=True)
df.round(3) 

Using the data from my dropbox folder, the round-function produces a whopping 7518 rows:

dfb = pd.read_pickle('dfna.pkl')
dfb.round(3)

This is strange. I solved it for now using this rather ugly line:

dfb = dfb.reset_index().round({'A': 1, 'B': 2, 'C': 3, 'D': 4}).set_index('tricile')

However, this is not ideal, given that pandas' round method acts in mysterious ways and may affect future programs.

[TomAugspurger]

Can you try making a simple, reproducible example that demonstrates the issue? http://matthewrocklin.com/blog/work/2018/02/28/minimal-bug-reports

We can't use pickle files for a unit test.

[TomAugspurger]

Actually, this looks like #21809

[blucap]

I posted the one on SO, then after jezrael commented there, I posted on git. Thanks for taking this forward!

…

On Mon, Jul 9, 2018 at 12:48 AM, Tom Augspurger ***@***.***> wrote: Actually, this looks like #21809 <#21809> — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#21810 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAlhkMBFzWpII1dEgPz2mhbSEdG6vKuDks5uEf-WgaJpZM4VGmSX> .

-- Kind regards, Martien

___________________________________ Martien Lubberink PhD | M: [email protected] | H: +64 4 232 9495 | M: +64 21 134 7500 | M: +316 8155 8570 | M: +44 774 558 4224 | No Need to Hit the 'Send' Key. Just Talk to Me.