CODEWITHSUNDEEP
pythonpandas
Pimin Konstantin Kefaloukos
Pimin Konstantin Kefaloukos

Reputation: 1580

Convert a pandas "Series of pair arrays" to a "two-column DataFrame"?

I have a Pandas Series that consists of arrays of pairs:

In [177]: pair_arrays
Out[177]: 
15192     [[1, 9], [2, 14], [4, 1], [5, 36], [6, 8], [7,...
16012     [[0, 107], [1, 42], [2, 22], [3, 59], [4, 117]...
17523     [[0, 44], [1, 36], [2, 43], [3, 28], [4, 52], ...
...

I would like to reshape that into a dataframe with two columns, 'x' and 'y', which has a shape similar to:

In [179]: pd.DataFrame([{'x':1, 'y':42}, {'x':4, 'y':12}], columns=['x', 'y'])
Out[179]: 
   x   y
0  1  42
1  4  12
...

How do I do this?

Upvotes: 2

Views: 2646

Answers (2)

TheBlackCat
TheBlackCat

Reputation: 10298

Assuming that each element in the series is an array of pairs, and each pair is a sequence, this should work:

pair_df = pd.DataFrame(np.vstack(pair_arrays.values), columns=['x','y'])

The key point is that pandas doesn't know how to work with object arrays. So what I am doing here is converting it to a numpy array of object arrays. Then I am stacking the object arrays, which gets you a 2D integer array, and then converting it back to a DataFrame.

Technically you don't currently need to use the values method to explicitly convert to a numpy array, but I think that is clearer and potentially safer long-term.

Upvotes: 3

Pimin Konstantin Kefaloukos
Pimin Konstantin Kefaloukos

Reputation: 1580

I can go via Python as follows:

pd.DataFrame(
  [item for sublist in pair_arrays.tolist() for item in sublist], 
  columns=['x', 'y']
)

This works for my use case, but maybe not ideal to go via Python like that.

Upvotes: 0

Related Questions