when to use iloc and loc for boolean

Question

I'm a bit confusing when using boolean series for indexing for pandas Dataframe. Should I use iloc or loc? or any better solution? for example

t1 = pd.DataFrame(np.ones([3,4]))
t1.iloc[1:3,0]=3

this line will give correct answer

t1.loc[:,(t1>2).any()]

but line with iloc wiill raise an error

t1.iloc[:,(t1>2).any()]

I check https://pandas.pydata.org/pandas-docs/stable/indexing.html, the page says both iloc and loc accept a boolean array. Why does iloc not work in my example? When to use iloc and loc? or are there any better alternatives?

Answer 1

The nuance is that iloc requires a Boolean array, while loc works with either a Boolean series or a Boolean array. The documentation is technically correct in stating that a Boolean array works in either case.

So, for iloc, extracting the NumPy Boolean array via pd.Series.values will work:

t1.iloc[:, (t1>2).any().values]