Serialize 'csv' file as binary and append to file

How is it possible to achieve the following at the same time in python 3:

Serialize column names and numerical data as a binary file
Reopen the file and append additional numerical data

For example with the following data:

import numpy as np

columns = ['a', 'b', 'c']
data = np.linspace(0, 1, num=10*3).reshape((10, 3))
data_for_appending = np.linspace(2, 3, num=10*3).reshape((10, 3))

My approach with numpy

This approach allows to save data and append additional data. However the column names are missing and loading requires several calls to np.load.

# storing the data
with open('out.npy', 'wb') as f:
    np.save(f, data)
    np.save(f, data_for_appending)

# loading the data
with open('out.npy', 'rb') as f:
    data1 = np.load(f)
    data2 = np.load(f)

My approach with pandas

This approach saves the data and header. However it seems not possible to append data to the file in a separate call.

import pandas as pd

df = pd.DataFrame(data, columns=columns)

# storing the data
df.to_pickle('out.pickle')

# loading the data
df2 = pd.read_pickle('out.pickle')

Solution

import pickle



# Write first df to pickle
data = {
    "name": ["Joe", "Mike", "Tony", "Susan"],
    "course": ["Masters", "Doctorate", "Graduate", "Bachelors"],
    "age": [27, 23, 21, 19],
}
df = pd.DataFrame(data)
df.to_pickle(path)

# Create new row df
new_row = {"name": "Phil", "course": "Associates", "age": 30}
new_row_df = pd.DataFrame(new_row, index=[0])
print(f"{new_row_df}\n")

# read original df from pickle
pickled_df = pd.read_pickle(path)

# concat dfs 
df_appended = pd.concat([new_row_df, pickled_df]).reset_index(drop=True)

# Dump concat df to pickle
with open(path, "wb") as f:
    pickle.dump(df_appended, f)

# read concat df from pickle
df = pd.read_pickle(path)
print(df)

You can append to the file without reading but the dfs wont be concatenated they are seperate entries. You can ofcourse read all the entries in a loop and concat later when it's time to read the file.

# Add new entries
with open(path, "ab") as f:
    pickle.dump(new_df, f)

# When ready to read and concat.
with open(path, "rb") as f:
    entries = []
    while True:
        try:
            entry = pickle.load(f)
        except EOFError:
            break
        entries.append(entry)

df = pd.concat(entries).reset_index(drop=True)
print(df)