This is a simple utility that conditionally splits a huge CSV file into smaller ones given some constraints. This is useful in many use-cases, but mine was that I wanted to examine/modify portions of the CSV file while not having to re-parse it or load it again in memory.

Remember, storing on a disk is much cheaper than just putting everything in RAM.

You will need Python 3.5+ or greater and I have developed it using Python 3.6. I also assume that you have docopt package installed, if not you will need to install it by using the following command:

pip3 install docopt

This command line utility has a lot of options that you can customize the splitting based on your preferences. It supports the following things:

• Splitting based on multiple column constraints
• Use an upper bound for iterations
• Use custom extension for your output files (different than .csv)
• Option to enable column name prepend against arguments.
• Option to suppress the first line naming (MatLab users rejoice! no more row offset!)
• Use different delimiters for input/output files
• Use a skip list that enables you to skip rows based on this

The general syntax of the command is this:

./csv_to_csvs.py split file.csv cons "[1, 2...,n]"

Where file.csv the file you want to split and the cons being the constraint list you want to have when you split the file; the constraint list has the following format:

"[idx1, idx2,...,idxn]"

As you might notice the constraint argument is an array of strings that has to be enclosed in " ". The constraints are encoded as a list of column label indexes starting from 0 and are expected to be in the order of enforcement, thus "[0, 1]" and "[1, 0]" are not the same.

Let's say you have the following column labels in csv file:

Index,Arrival_Time,Creation_Time,x,y,z,User,Model,Device,gt

Now say you want to get file splits based on each User, what you could do is:

./csv_to_csvs.py split myfile.csv cons "[6]"

Now you would get a number of csv files containing each information respective to each unique user found during execution.

Now let's take it a step further, say you want to create the csv files based on each Device and User, that could the achieved like so:

./csv_to_csvs.py split myfile.csv cons "[6, 8]"

Do note that the User has precedence over the Device, should you want it to be the other way around then you would do:

./csv_to_csvs.py split myfile.csv cons "[8, 6]"

On my 2016 Macbook Pro I can parse and split CSV files at a rate of around 35-25 MB/s; but do note that YMMV.

Please note that if you encounter while executing the program in Mac OS an error message saying that you have too many file streams open, then you should adjust this limit by running ulimit -n <value> command inside the terminal.

Unless otherwise noted, this work is licensed under the terms and conditions of GPLv3.