How to explode a list inside a Dataframe cell into separate rows

Question

I'm looking to turn a pandas cell containing a list into rows for each of those values.

So, take this:

If I'd like to unpack and stack the values in the nearest_neighbors column so that each value would be a row within each opponent index, how would I best go about this? Are there pandas methods that are meant for operations like this?

Answer 1

Exploding a list-like column has been simplified significantly in pandas 0.25 with the addition of the explode() method:

df = (pd.DataFrame({'name': ['A.J. Price'] * 3, 
                    'opponent': ['76ers', 'blazers', 'bobcats'], 
                    'nearest_neighbors': [['Zach LaVine', 'Jeremy Lin', 'Nate Robinson', 'Isaia']] * 3})
      .set_index(['name', 'opponent']))

df.explode('nearest_neighbors')

Out:

                    nearest_neighbors
name       opponent                  
A.J. Price 76ers          Zach LaVine
           76ers           Jeremy Lin
           76ers        Nate Robinson
           76ers                Isaia
           blazers        Zach LaVine
           blazers         Jeremy Lin
           blazers      Nate Robinson
           blazers              Isaia
           bobcats        Zach LaVine
           bobcats         Jeremy Lin
           bobcats      Nate Robinson
           bobcats              Isaia

Answer 2

Instead of using apply(pd.Series) you can flatten the column. This improves performance.

df = (pd.DataFrame({'name': ['A.J. Price'] * 3, 
                'opponent': ['76ers', 'blazers', 'bobcats'], 
                'nearest_neighbors': [['Zach LaVine', 'Jeremy Lin', 'Nate Robinson', 'Isaia']] * 3})
  .set_index(['name', 'opponent']))



%timeit (pd.DataFrame(df['nearest_neighbors'].values.tolist(), index = df.index)
           .stack()
           .reset_index(level = 2, drop=True).to_frame('nearest_neighbors'))

1.87 ms ± 9.74 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)


%timeit (df.nearest_neighbors.apply(pd.Series)
          .stack()
          .reset_index(level=2, drop=True)
          .to_frame('nearest_neighbors'))

2.73 ms ± 16.7 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

Answer 3

I think this a really good question, in Hive you would use EXPLODE, I think there is a case to be made that Pandas should include this functionality by default. I would probably explode the list column with a nested generator comprehension like this:

pd.DataFrame({
    "name": i[0],
    "opponent": i[1],
    "nearest_neighbor": neighbour
    }
    for i, row in df.iterrows() for neighbour in row.nearest_neighbors
    ).set_index(["name", "opponent"])

Answer 4

The fastest method I found so far is extending the DataFrame with .iloc and assigning back the flattened target column.

Given the usual input (replicated a bit):

df = (pd.DataFrame({'name': ['A.J. Price'] * 3, 
                    'opponent': ['76ers', 'blazers', 'bobcats'], 
                    'nearest_neighbors': [['Zach LaVine', 'Jeremy Lin', 'Nate Robinson', 'Isaia']] * 3})
      .set_index(['name', 'opponent']))
df = pd.concat([df]*10)

df
Out[3]: 
                                                   nearest_neighbors
name       opponent                                                 
A.J. Price 76ers     [Zach LaVine, Jeremy Lin, Nate Robinson, Isaia]
           blazers   [Zach LaVine, Jeremy Lin, Nate Robinson, Isaia]
           bobcats   [Zach LaVine, Jeremy Lin, Nate Robinson, Isaia]
           76ers     [Zach LaVine, Jeremy Lin, Nate Robinson, Isaia]
           blazers   [Zach LaVine, Jeremy Lin, Nate Robinson, Isaia]
...

Given the following suggested alternatives:

col_target = 'nearest_neighbors'

def extend_iloc():
    # Flatten columns of lists
    col_flat = [item for sublist in df[col_target] for item in sublist] 
    # Row numbers to repeat 
    lens = df[col_target].apply(len)
    vals = range(df.shape[0])
    ilocations = np.repeat(vals, lens)
    # Replicate rows and add flattened column of lists
    cols = [i for i,c in enumerate(df.columns) if c != col_target]
    new_df = df.iloc[ilocations, cols].copy()
    new_df[col_target] = col_flat
    return new_df

def melt():
    return (pd.melt(df[col_target].apply(pd.Series).reset_index(), 
             id_vars=['name', 'opponent'],
             value_name=col_target)
            .set_index(['name', 'opponent'])
            .drop('variable', axis=1)
            .dropna()
            .sort_index())

def stack_unstack():
    return (df[col_target].apply(pd.Series)
            .stack()
            .reset_index(level=2, drop=True)
            .to_frame(col_target))

I find that extend_iloc() is the fastest:

%timeit extend_iloc()
3.11 ms ± 544 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

%timeit melt()
22.5 ms ± 1.25 ms per loop (mean ± std. dev. of 7 runs, 100 loops each)

%timeit stack_unstack()
11.5 ms ± 410 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

Answer 5

Extending Oleg's .iloc answer to automatically flatten all list-columns:

def extend_iloc(df):
    cols_to_flatten = [colname for colname in df.columns if 
    isinstance(df.iloc[0][colname], list)]
    # Row numbers to repeat 
    lens = df[cols_to_flatten[0]].apply(len)
    vals = range(df.shape[0])
    ilocations = np.repeat(vals, lens)
    # Replicate rows and add flattened column of lists
    with_idxs = [(i, c) for (i, c) in enumerate(df.columns) if c not in cols_to_flatten]
    col_idxs = list(zip(*with_idxs)[0])
    new_df = df.iloc[ilocations, col_idxs].copy()

    # Flatten columns of lists
    for col_target in cols_to_flatten:
        col_flat = [item for sublist in df[col_target] for item in sublist]
        new_df[col_target] = col_flat

    return new_df

This assumes that each list-column has equal list length.

Answer 6

Similar to Hive's EXPLODE functionality:

import copy

def pandas_explode(df, column_to_explode):
    """
    Similar to Hive's EXPLODE function, take a column with iterable elements, and flatten the iterable to one element 
    per observation in the output table

    :param df: A dataframe to explod
    :type df: pandas.DataFrame
    :param column_to_explode: 
    :type column_to_explode: str
    :return: An exploded data frame
    :rtype: pandas.DataFrame
    """

    # Create a list of new observations
    new_observations = list()

    # Iterate through existing observations
    for row in df.to_dict(orient='records'):

        # Take out the exploding iterable
        explode_values = row[column_to_explode]
        del row[column_to_explode]

        # Create a new observation for every entry in the exploding iterable & add all of the other columns
        for explode_value in explode_values:

            # Deep copy existing observation
            new_observation = copy.deepcopy(row)

            # Add one (newly flattened) value from exploding iterable
            new_observation[column_to_explode] = explode_value

            # Add to the list of new observations
            new_observations.append(new_observation)

    # Create a DataFrame
    return_df = pandas.DataFrame(new_observations)

    # Return
    return return_df

Answer 7

Here is a potential optimization for larger dataframes. This runs faster when there are several equal values in the "exploding" field. (The larger the dataframe is compared to the unique value count in the field, the better this code will perform.)

def lateral_explode(dataframe, fieldname): 
    temp_fieldname = fieldname + '_made_tuple_' 
    dataframe[temp_fieldname] = dataframe[fieldname].apply(tuple)       
    list_of_dataframes = []
    for values in dataframe[temp_fieldname].unique().tolist(): 
        list_of_dataframes.append(pd.DataFrame({
            temp_fieldname: [values] * len(values), 
            fieldname: list(values), 
        }))
    dataframe = dataframe[list(set(dataframe.columns) - set([fieldname]))]\ 
        .merge(pd.concat(list_of_dataframes), how='left', on=temp_fieldname) 
    del dataframe[temp_fieldname]

    return dataframe

Answer 8

Use apply(pd.Series) and stack, then reset_index and to_frame

In [1803]: (df.nearest_neighbors.apply(pd.Series)
              .stack()
              .reset_index(level=2, drop=True)
              .to_frame('nearest_neighbors'))
Out[1803]:
                    nearest_neighbors
name       opponent
A.J. Price 76ers          Zach LaVine
           76ers           Jeremy Lin
           76ers        Nate Robinson
           76ers                Isaia
           blazers        Zach LaVine
           blazers         Jeremy Lin
           blazers      Nate Robinson
           blazers              Isaia
           bobcats        Zach LaVine
           bobcats         Jeremy Lin
           bobcats      Nate Robinson
           bobcats              Isaia

Details

In [1804]: df
Out[1804]:
                                                   nearest_neighbors
name       opponent
A.J. Price 76ers     [Zach LaVine, Jeremy Lin, Nate Robinson, Isaia]
           blazers   [Zach LaVine, Jeremy Lin, Nate Robinson, Isaia]
           bobcats   [Zach LaVine, Jeremy Lin, Nate Robinson, Isaia]

Answer 9

In the code below, I first reset the index to make the row iteration easier.

I create a list of lists where each element of the outer list is a row of the target DataFrame and each element of the inner list is one of the columns. This nested list will ultimately be concatenated to create the desired DataFrame.

I use a lambda function together with list iteration to create a row for each element of the nearest_neighbors paired with the relevant name and opponent.

Finally, I create a new DataFrame from this list (using the original column names and setting the index back to name and opponent).

df = (pd.DataFrame({'name': ['A.J. Price'] * 3, 
                    'opponent': ['76ers', 'blazers', 'bobcats'], 
                    'nearest_neighbors': [['Zach LaVine', 'Jeremy Lin', 'Nate Robinson', 'Isaia']] * 3})
      .set_index(['name', 'opponent']))

>>> df
                                                    nearest_neighbors
name       opponent                                                  
A.J. Price 76ers     [Zach LaVine, Jeremy Lin, Nate Robinson, Isaia]
           blazers   [Zach LaVine, Jeremy Lin, Nate Robinson, Isaia]
           bobcats   [Zach LaVine, Jeremy Lin, Nate Robinson, Isaia]

df.reset_index(inplace=True)
rows = []
_ = df.apply(lambda row: [rows.append([row['name'], row['opponent'], nn]) 
                         for nn in row.nearest_neighbors], axis=1)
df_new = pd.DataFrame(rows, columns=df.columns).set_index(['name', 'opponent'])

>>> df_new
                    nearest_neighbors
name       opponent                  
A.J. Price 76ers          Zach LaVine
           76ers           Jeremy Lin
           76ers        Nate Robinson
           76ers                Isaia
           blazers        Zach LaVine
           blazers         Jeremy Lin
           blazers      Nate Robinson
           blazers              Isaia
           bobcats        Zach LaVine
           bobcats         Jeremy Lin
           bobcats      Nate Robinson
           bobcats              Isaia

EDIT JUNE 2017

An alternative method is as follows:

>>> (pd.melt(df.nearest_neighbors.apply(pd.Series).reset_index(), 
             id_vars=['name', 'opponent'],
             value_name='nearest_neighbors')
     .set_index(['name', 'opponent'])
     .drop('variable', axis=1)
     .dropna()
     .sort_index()
     )

Answer 10

Nicer alternative solution with apply(pd.Series):

df = pd.DataFrame({'listcol':[[1,2,3],[4,5,6]]})

# expand df.listcol into its own dataframe
tags = df['listcol'].apply(pd.Series)

# rename each variable is listcol
tags = tags.rename(columns = lambda x : 'listcol_' + str(x))

# join the tags dataframe back to the original dataframe
df = pd.concat([df[:], tags[:]], axis=1)