Pandas .drop() reacts differently when lables changed

panads version is 0.20.2

df=pd.DataFrame({'0':['x',np.nan,np.nan],'col':range(3)})
     0  col
0    x    0
1  NaN    1
2  NaN    2

df.set_index('0',inplace=True)
df
     col
0       
x      0
NaN    1
NaN    2

#df.drop(df.index[[0]])  #this can drop  reminded by TomAugspurger 
df.drop(df.index[[1]])  #this cannot
     col
0       
x      0
NaN    1
NaN    2

df.drop(df.index[[0,1]]) #drop all 
Empty DataFrame
Columns: [col]
Index: []

and if replace NaN(which is np.nan ) in the df with float('nan') , it will get other weirder results

could you please tell me what leads to the problem above? Thx

Comment From: TomAugspurger

Your first one seems OK to me on master, could you try upgrading your version of pandas?

In [10]: df=pd.DataFrame({'0':['x',np.nan,np.nan],'col':range(3)}).set_index('0')

In [11]: df.drop(df.index[0])
Out[11]:
     col
0
NaN    1
NaN    2

For your second issue, this probably comes from NaN != NaN. I'd recommend not putting NaNs in the index.

and when mix float('nan') with np.nan , will get other weirder results

What do you mean?

Comment From: TomAugspurger

Let us know if you have any other questions.