CSV vs MAT files

Should I store very large amount of data as .mat files or .csv. Which is more: (1) efficient when it comes to reading the data (2) more compressed in terms of size

[Jason Ross]

Some questions:

- What is "a very large amount of data"? Each of us has our own definition ... some people might think in MB, others in GB, some in TB and some in PB.

- Is this a collection of large files or a collection of small files?

- Is this stored in a directory hierarchy or a collection of files?

- Will other applications have to access the data, or only MATLAB?

- Is this stored locally on the PC or on a network location?

[Sean de Wolski]

Whoever here is think in PB you win! I think TBs are big.

[Jason Ross]

Come on, a 2TB drive is $100 nowadays :). Of course, that's not enterprise class storage but it's fairly easy to fill up with virtual machine images and HD video (in a DVR, for instance). Or piles of data coming going into or out of something or other.

[Sean de Wolski]

We take high resolution xmt scans, do all sorts of fun math on them, and we may fill a TB over the course of a year.

I do enjoy the reward of knowing that it is faster and (obviously) requires less space to recompute something than to store it and load it.

Answer 1

[Ben Mitch]

I think the answer to both (1) and (2) is .mat file. ASCII files (like CSV) require conversion to and from the format in memory (binary), which makes them slow. Moreover, if written at more than 6 significant figures, they are bigger than the usual (double precision) binary format as well. Therefore, you should probably only use CSV if either (a) you need to exchange data with software that can read CSV but cannot read MAT (like Excel) or (b) you want to be able to peruse the data yourself in a text editor or CSV editor.

Aside, if speed is your absolute goal, consider using

save myvariable myfile -v6

because both the save() and load() commands are much quicker if compression is disabled like this (compression was not available in Version 6 of Matlab). Vice versa, if small file size is your goal, use the usual save/load commands.

Depending on the data, you might get some additional savings by first casting to a smaller data type. For instance, if you are storing data as double precision but are confident that single precision will be enough... try this:

a = randn(1000, 1);
save a1 a
a = single(a);
save a2 a

and check the filesizes. Note that the filesize has got smaller in a2 because you've thrown away information which you judged yourself to be irrelevant.

[Ken Atwell]

Another advantage of a MAT file is random access -- a CSV file does not necessarily have predictable line lengths, so you cannot reliably seek into the middle of the file. When working with big data, the ability to seek and incrementally read is useful. More at:

http://blogs.mathworks.com/loren/2011/10/14/new-mat-file-functionality-in-r2011b/

Answer 2

[Sean de Wolski]

• If you're going to be using it outside of MATLAB -> CSV
• If you're only using it within MATLAB and you're motivating to writing/reading directly, then write it to binary using fwrite, and fread to pull it in. I used to do this, but got lazy and realized it's much easier even if slower to use MAT files
• If you're lazy and want something easy/the ability to store multiple variables at once -> .MAT

My suggestion after years of being angry at reading in a binary file with the wrong dimensions: use .mat files.

Answer 3

[Sam Da]

I am going to use the .mat files both in Matlab and R. R does have packages to load the .mat files. So, is .mat preferred or .csv preferred for speed and size of execution.

I am talking about let's say 500GB of data.

[Sean de Wolski]

Do you have 500GB of RAM? You won't be storing all 500GB in one mat file right? As long as your mat files don't exceed at least half the RAM you have, you should be fine. You can store more than that, but you won't be able to do anything with it without exceeding your RAM limit.

[Walter Roberson]

You can append to existing .mat files, if you are adding new variables. (You cannot append to an existing variable except by rewriting the whole variable.)

Answer 4

[Lisa Justin]

i have a csv file and want to convert it to a matlab file. how can i do this?

[Sean de Wolski]

Lisa, Please ask a new question.