Document that we don't support the compatibility with non-Koalas APIs yet.

Seems like people want to convert their codes directly from pandas to Koalas. One case I often observe is, they want to convert the codes that works together with other Python standard functions such as max, min, or list/generator comprehensions, e.g.)

import pandas as pd
data = []
for a in pd.Series([1, 2, 3]):
    data.append(a)

pd.DataFrame(data)

In Koalas, such example does not work. We should preemptively document and guide users to stick to Koalas APIs only.

[HyukjinKwon]

I think we should better move them to Best Practice. I think we could rephrase, for example as below. Feel free to reword or rephrase.

Title: Use Koalas APIs directly whenever possible

Contents: While Koalas has similar APIs with pandas, some APIs are not explicitly supported. For example, Python built-in functions such as min, max, etc. require the given argument to be iterrable. Koalas does not implement __iter__() yet to prevent users to collect all data into the client (driver) side from the cluster. See the example below:

>>> import pandas as pd
>>> max(pd.Series([1, 2, 3]))
3
>>> min(pd.Series([1, 2, 3]))
1
>>> sum(pd.Series([1, 2, 3]))
6

pandas dataset live in the local, iterable … blah blah …

>>> import databricks.koalas as ks
>>> ks.Series([1, 2, 3]).max()
3
>>> ks.Series([1, 2, 3]).min()
1
>>> ks.Series([1, 2, 3]).sum()
6

Koalas performes it in a distributed manner… blah blah

… Another common pattern from pandas users is to rely on list or generator comprehensions …:

>>> import pandas as pd
>>> data = []
>>> pser = pd.Series([1, 2, 3])
>>> for v in pser:
...     data.append(v + 1)
>>> pd.Series(data)

In Koalas, you can do it via:

>>> import databricks.koalas as ks
>>> kser = ks.Series([1, 2, 3])
>>> kser + 1  # or kser.apply example? or kdf.map_in_pandas example? 

In case of NumPy universial functions, they are supported and can be naturally used in most cases. -> it was added https://github.com/databricks/koalas/pull/1096 https://github.com/databricks/koalas/pull/1106 https://github.com/databricks/koalas/pull/1128 FYI

Using to_numpy should still be discouraged and the last resort.

[beobest2]

okay~ I’ll open a PR. thank you