Is it possible to add columns to a pandas dataframe without filling it with any values?

So I have a pandas dataframe which is being passed from function to function. However a the moment I do not have any data to populate the rows with.

Furthermore, because of the way the code is structured, the dataframe needs to have certain columns.

Is it possible to add columns to a dataframe without mapping it to any value? I also don't want to map it to 0 or None or any default value. I would just like the empty dataframe with certain columns.

e.g.

...

def _trades(self, trades_df):

    trades_df = trades_df.rename(columns={'timestamp': 'trade_timestamp'})
    trades_df['publication_timestamp'] = trades_df['trade_timestamp']
    trades_df['trade_id'] = trades_df['trade_id'].astype(str)


    # set printable column - this way is empty dataframe safe
    trades_df['printable'] = True

    # No trade_types to map explicitly
    trades_df['trade_type'] = None

    trades_df['implied'] = 0

    return trades_df

As you can see above the implied column is mapped to 0 and trade_type is also mapped to None.

However I just want to add the columns without mapping it with any default value.

In pandas, the dataframe object is tabular. This means it contains a rectangular collection of values. This rectangle can have zero rows, in which case columns can be added without any values in those columns.

However, if the rectangle has a non-zero number of rows, then each row in a column must have a value. This value can be None (python's null object value) or NaN (numpy's not-a-number value) or the empty string, or even an empty python sequence (tuple or list). But there is no such thing, in a dataframe with both axes (rows and columns) having non-zero length, as a cell without any value.

The one other thing you can do is to initialize the data in a new column using numpy.empty() which according to the docs will:

> Return a new array of given shape and type, without initializing entries.

Consider this code:

trades_df['trade_type'] = np.empty([len(trades_df)])
trades_df['implied'] = np.empty([len(trades_df)])

Input:

   trade_timestamp  publication_timestamp trade_id  printable
0                1                      1      101       True
1                2                      2      102       True
2                3                      3      103       True

Output:

   trade_timestamp  publication_timestamp trade_id  printable     trade_type
0                1                      1      101       True  6.953347e-310
1                2                      2      102       True  6.953347e-310
2                3                      3      103       True  6.953347e-310
   trade_timestamp  publication_timestamp trade_id  printable     trade_type        implied
0                1                      1      101       True  6.953347e-310  1.232637e-311
1                2                      2      102       True  6.953347e-310  1.232637e-311
2                3                      3      103       True  6.953347e-310  1.232637e-311

Th above example passes the default dtype argument float to numpy.empty(), but it is possible to use other numpy scalar types instead.

yes, of course:

df["C"] = ""