ENH: Make concat succeed on empty list.

[frobnitzem]

Is your feature request related to a problem?

Yes, pd.concat([]) currently raises the exception, ValueError("No objects to concatenate").

Describe the solution you'd like

To be correct on all input sizes, it should return an empty DataFrame.

API breaking implications

Obviously, concat will no longer detect empty data. However, empty data will detect itself.

Additional context

The concatenation operation is for lists what addition is for numbers. Concatenation of an empty list should produce an empty dataframe, because the empty dataframe represents 0. Consider

sum([])

which python evaluates to 0.

This makes concatenation and sum commute, sum([1] + []) == sum([1]) + sum([]).

This already works for an empty dataframe:

pd.concat([pd.DataFrame({'A':[1]}), pd.DataFrame()])

if is very odd for concat([]) not to produce a DataFrame, since its input has the correct type.

[ivanovmg]

I guess it is not clear what should be the return type.
If we concat dataframes, then the function returns a dataframe.
Similarly, for series objects it returns a series.
However, what should we return in case of an empty input?

[attack68]

I guess it is not clear what should be the return type.
If we concat dataframes, then the function returns a dataframe.
Similarly, for series objects it returns a series.
However, what should we return in case of an empty input?

I agree, and if Series is to be interpreted as a 'simpler' object (i.e. df["colA"] returns a Series not a DataFrame), then arguably one might suggest returning a Series rather than a DataFrame, if anything at all is returned.

[frobnitzem]

Sounds like a solution would be for concat([]) to return a new object type representing both empty Series and empty DataFrames. Perhaps there's a base class that makes sense there - I don't know enough to say.

For now, I just create my own concatDF function that only works on DataFrames.

[yushao2]

Sounds like a solution would be for concat([]) to return a new object type representing both empty Series and empty DataFrames. Perhaps there's a base class that makes sense there - I don't know enough to say.

For now, I just create my own concatDF function that only works on DataFrames.

Here's an idea, incorporating an optional initializer for the pd.concat function, similar to reduce:

https://docs.python.org/3/library/functools.html#functools.reduce

Otherwise, it seems like the base class is generic.NDFrame:

pandas/pandas/core/generic.py

Line 187 in 3cce96f

class NDFrame(PandasObject, indexing.IndexingMixin):

[yushao2]

Sounds like a solution would be for concat([]) to return a new object type representing both empty Series and empty DataFrames. Perhaps there's a base class that makes sense there - I don't know enough to say.

For now, I just create my own concatDF function that only works on DataFrames.

I created a PR for this for the first solution mentioned -- feel free to take a look and share your thoughts.

It's my first time contributing to this project so I'm sorry if I've made any mistakes.

Thanks!

[mroeschke]

I find the current behavior correct as is.

For reference np.concatenate also raises a similar error as pd.concat

In [1]: np.concatenate([])
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-1-7806ca832c55> in <module>
----> 1 np.concatenate([])

<__array_function__ internals> in concatenate(*args, **kwargs)

ValueError: need at least one array to concatenate

And I believe it's more explicit for the user to decide what to do in code flow if they get to a pd.concat([]) situation.

So I'd be -1 on any changes here.