In this post I am taking a closer look at tables and strings in matlab.

Matlab continuously evolves, and the two data types are fairly recent additions to the language: table was added in R2013b and string in R2016b. Table arrays store tabular data and can make its handling (sorting, filtering, joining etc.) much more convenient than “classical” cell arrays. In a way, they are similar to dataframes known from R or python / pandas, although more basic. String arrays are a proper string type, and in my opinion they are much more workable then their predecessors, character arrays and cellstrings.

To demonstrate how useful they can be together, I will show an example modelled after preparing an input for SPM batch (specifically, second level two sample t-test design). If you work in neuroimaging, you’ll probably be familiar with SPM, but if not, you will be able to follow along just as well.

Problem statement

What we have is a tab-separated file participants.tsv with a list of subjects (in accordance with the BIDS specification):

participant_id age group
sub-01         34  read
sub-02         12  write
sub-03         33  read

We also have a set of intermediate files (first level output), which we organised in the following way (this time according to our whim, rather than any standard):

/work/sub-01/task-motor/con_0001.nii
/work/sub-01/task-nback/con_0001.nii
/work/sub-02/task-motor/con_0001.nii
/work/sub-02/task-nback/con_0001.nii

What we want to get is two lists of files for the motor task: one for the read group, and one for the write group:

group1 = {
	/work/sub-01/task-motor/con_0001.nii;
	/work/sub-03/task-motor/con_0001.nii;
	...
	}
group2 = {
	/work/sub-02/task-motor/con_0001.nii;
	...
	}

Solution

Let’s start by reading the participant file:

participants = readtable('/path/to/participants.tsv', ...
	'FileType', 'text', 'Delimiter', '\t', 'TextType', 'string')

Unfortunately, readtable recognizes .txt, .dat or .csv extensions, but not .tsv, so we have to specify both FileType and Delimiter (either \t or tab will work) in addition to the obligatory file name. Additionally, we use TextType to import text data as string arrays, rather than character vectors. This produces the following:

participants =

  3×3 table

    participant_id    age     group 
    ______________    ___    _______

       "sub-01"       34     "read" 
       "sub-02"       12     "write"
       "sub-03"       33     "read" 

Next, let’s add another column with paths to the con files:

participants.con_files = fullfile(...
	"/work", participants.participant_id, "task-motor", "con_0001.nii")

The fullfile function joins its arguments by adding a path separator (/ or \, depending on system). One of the inputs we gave, particiants.participant_id is a string array (with as many rows as there are subjects), and consequently the output will be a string array of the same shape. Adding a column is as simple as retrieving one, and can be done by using a dot notation with a new name. Convenient. This is what the table looks as a result:

participants =

  3×4 table

    participant_id    age     group                   con_files               
    ______________    ___    _______    ______________________________________

       "sub-01"       34     "read"     "/work/sub-01/task-motor/con_0001.nii"
       "sub-02"       12     "write"    "/work/sub-02/task-motor/con_0001.nii"
       "sub-03"       33     "read"     "/work/sub-03/task-motor/con_0001.nii"

We can next filter the table for each group in turn.

read_con = participants{participants.group == "read", "con_files"}
write_con = participants{participants.group == "write", "con_files"}

Here, we used brace indexing to get a subset of rows (since the group column is a string array, we could use == to obtain a logical array of a matching shape for logical indexing) and a single column (by giving its name). In a typical matlab fashion, using round brackets would keep the type of the indexed object and return a table with selected rows and columns, while the curly brackets used above return column content¹ (in this case, a string array). This produces:

read_con = 

  2×1 string array

    "/work/sub-01/task-motor/con_0001.nii"
    "/work/sub-03/task-motor/con_0001.nii"


write_con = 

    "/work/sub-02/task-motor/con_0001.nii"

Finally, since SPM batch requires cellstrings rather than string arrays, we can do a simple conversion:

group1 = cellstr(read_con)
group2 = cellstr(write_con)

group1 =

  2×1 cell array

    {'/work/sub-01/task-motor/con_0001.nii'}
    {'/work/sub-03/task-motor/con_0001.nii'}


group2 =

  1×1 cell array

    {'/work/sub-02/task-motor/con_0001.nii'}

Further comments

Naturally, the example above was fairly basic, and there is more that can be done with strings and tables. Here are some things that I found particularly useful.

You can join string arrays using + (eg. you could use "sub-" + participants.participant_id if the “sub-” prefix was missing). If you are a fan of sprintf, you can use compose as its “vectorised” alternative (eg. compose("sub-%s", participants.participant_id)) for a similar effect. Should you wish to replace part of the string, you can use things like replace, eg. replace(participants.participant_id, "sub-", "subject-"). For filtering, you might want to use startsWith and the like.

For tables, in addition to the regular size function there are intuitive height and width functions. If you have another table (perhaps with behavioural scores) with the same participant IDs but different ordering, you can join them using join. And finally, if you need to sort a table first by one column, and then by another (breaking ties), you can do so using sortrows, as in sortrows(participants, {'group', 'age'}) (which also allows you to specify ascending or descending order independently).

Footnotes