Get a subset of columns from a timeseries object as a timeseries

I have a timeseries containing data for 4 variables:

Ts = 0.2; nT = 50;
t = Ts*(0:nT-1)';
u1 = 50*idinput(nT);
u2 = 50*idinput(nT);
u3 = 50*idinput(nT);
u4 = 50*idinput(nT);
u = timeseries([u1 u2 u3 u4],t);

I see there is a function called getsamples to get a subset of the data samples.

E.g. to get the first data point:

>> getsamples(u,1)
  timeseries

  Common Properties:
            Name: 'unnamed'
            Time: [1x1 double]
        TimeInfo: [1x1 tsdata.timemetadata]
            Data: [1x4 double]
        DataInfo: [1x1 tsdata.datametadata]

  More properties, Methods
  

But how do I get a subset of the columns as a timeseries?

  u1 = ...  % ?

E.g. what if I want to plot just u1?

Obviously, I can do this:

  plot(u.Time,u.Data(:,1))

But isn't there a simpler way similar to how I would plot a timeseries with only one variable?

plot(u(:,1))  % indexing like this does not work

[Bill Tubbs]

Just noticed there is a new function called ts2timetable introduced in version 2021a which might be usefull for converting multiple timeseries objects into a timetable.

Answer 1

[dpb]

plot(u.Time,u.Data(:,1))

"But isn't there a simpler way similar to how I would plot a timeseries with only one variable?"

Once you made a timeseries object out of a multi-dimensional array, no. The .Data property is a 2D array and so you must subscript the array to refer to only a single column (or subset of columns) just like any other array.

I've found the timeseries object to be more cumbersome than the features it offers owing to the way interaction with it is built; I'd probably just revert to a regular timetable instead, unless you do have the one or two specific things the timeseries supports like the event.

But, the object is constructed such that all the data associated with it is either a vector or a 2D array, it doesn't have the concept of there being separate but associated timeseries variables. That can only be done in a timetable

As noted, at least so far I've failed to find sufficient added features in the timeseries to be able to have found a use for it for either anything I've tried to do myself or even to answer any Q? posed here (other than just syntax issues like this one that (almost?) all of which have resulted in suggesting other solutions are more easily implemented than by continuing to fight the timeseries limitations/design).

[Bill Tubbs]

Thanks for the advice. I'm new to Matlab and trying to use high-level object types rather than working directly with arrays. What I'm really looking for is something like Python Pandas with its integrated Series and Dataframe objects. I'll checkout timetable. Seems like it might be the equivalent of a Dataframe with a datetime index. However, it doesn't look like Simulink 'to workspace' blocks have the option of outputting a timetable. Maybe I can construct one afterwards from a collection of timeseries. Although, it looks like you can log signals in timetable format.

[dpb]

I "know nuthink!" of Python but the example at the above link

df2 = pd.DataFrame(
   ...:     {
   ...:         "A": 1.0,
   ...:         "B": pd.Timestamp("20130102"),
   ...:         "C": pd.Series(1, index=list(range(4)), dtype="float32"),
   ...:         "D": np.array([3] * 4, dtype="int32"),
   ...:         "E": pd.Categorical(["test", "train", "test", "train"]),
   ...:         "F": "foo",
   ...:     }
   ...: )
   ...: 

In [10]: df2
Out[10]: 
     A          B    C  D      E    F
0  1.0 2013-01-02  1.0  3   test  foo
1  1.0 2013-01-02  1.0  3  train  foo
2  1.0 2013-01-02  1.0  3   test  foo
3  1.0 2013-01-02  1.0  3  train  foo

looks just like a MATLAB table object with slightly differing syntax and that the default display of a a table in MATLAB doesn't show a row index (altho you can define one if wanted).

>> df2=table(ones(4,1), ...
             repmat(datetime("20130102",'InputFormat','yyyyddMM','format','yyyy-MM-dd'),4,1), ...
             single(ones(4,1)), ...
             int32(3*ones(4,1)), ...
             categorical(["test", "train", "test", "train"].'), ...
             repmat("foo",4,1), ...
             'variablenames',cellstr(['A':'F'].'))
>> df2
df2 =
  4×6 table
    A        B         C    D      E        F  
    _    __________    _    _    _____    _____
    1    2013-01-02    1    3    test     "foo"
    1    2013-01-02    1    3    train    "foo"
    1    2013-01-02    1    3    test     "foo"
    1    2013-01-02    1    3    train    "foo"
>> 

Other than MATLAB shows integral values of floating point variables in the command window as integers owing to using a '%g' format string by default and differentiates between char(), cellstr() and string() outputs by using '', ,{''} or "", respectively, the result is identical in content and in holding disparate data types in the various columns.

[Bill Tubbs]

Yeah, tables are pretty good and I started using them first (great for saving/loading csv files). Unfortunately the indexing capabilities are quite limit (only character vectors are allowed for the RowNames property). That's why I am now looking to use timeseries or timetables instead.

Pandas dataframes have some pretty sophisticated indexing cabilities (including DatetimeIndex and multi-level indexes) so I'm not really comparing tables with them.

But as a simple example, how would I do this with a Matlab table:

selected_data = df.loc['2013-01-02 00:00':'2013-01-02 12:00', ['y1','y2','y5']]

[Steven Lord]

To index into a tabular array using dates and times, use a timetable instead of a table.

rng default
N = datetime('now')

N = datetime
   04-Apr-2021 21:37:08

% Random times between an hour before and an hour after right now
dt = N + minutes(randi([-60 60], 10, 1));

TT = timetable(dt, (1:10).')

TT = 10×1 timetable
             dt             Var1
    ____________________    ____

    04-Apr-2021 22:15:08      1 
    04-Apr-2021 22:26:08      2 
    04-Apr-2021 20:52:08      3 
    04-Apr-2021 22:27:08      4 
    04-Apr-2021 21:53:08      5 
    04-Apr-2021 20:48:08      6 
    04-Apr-2021 21:10:08      7 
    04-Apr-2021 21:43:08      8 
    04-Apr-2021 22:32:08      9 
    04-Apr-2021 22:33:08     10 

% Get rows between [now + 15 minutes, now + 45 minutes]
tr = timerange(N+minutes(15), N+minutes(45), 'closed')

tr = 
	timetable timerange subscript:

		Select timetable rows with times in the closed interval:
		[04-Apr-2021 21:52:08, 04-Apr-2021 22:22:08]

	See Select Timetable Data by Row Time and Variable Type.

TT(tr, :)

ans = 2×1 timetable
             dt             Var1
    ____________________    ____

    04-Apr-2021 22:15:08     1  
    04-Apr-2021 21:53:08     5  

[Bill Tubbs]

Great, thanks. Sounds like we're all agreed. timetable is the solution!

[dpb]

As Steven notes and shows, MATLAB has both the timetable and the table instead of combining the functionality. There are some restrictions that the timerange only works with timetables, but that doesn't mean there isn't syntax to let you do the same thing in a table -- only that the syntax is a little different.

With the same table above and on the presumption that the y in the ['y1','y2','y5'] syntax is referring to a generic y as variables by position,

>> df2(iswithin(df2.B,datetime(2013,1,2),datetime(2013,1,2,12,0,0)),[1,2,5])
ans =
  4×3 table
    A           B              E  
    _    ________________    _____
    1    2013-01-02 00:00    test 
    1    2013-01-02 04:00    train
    1    2013-01-02 08:00    test 
    1    2013-01-02 12:00    train
>> 

after having augmented the initial datetime variable B with [hours(0:4:12)].'

If, the example came from another table with variables y1, y2, ...yn then the equivalent in ML is:

df2(iswithin(df2.B,datetime(2013,1,2),datetime(2013,1,2,12,0,0)),{'A','B','E'})

The same level of addressing is available; when combined with rowfun and/or varfun and grouping variables, there's little one can think of that isn't accomplished pretty easily, albeit there is some learning time to become familiar with the particulars.

[dpb]

ADDENDUM:

iswithin is my little helper function to move the logical test to a higher level of abstraction --

function flg=iswithin(x,lo,hi)
% returns T for values within range of input
% SYNTAX:
%  [log] = iswithin(x,lo,hi)
%      returns T for x between lo and hi values, inclusive

  flg= (x>=lo) & (x<=hi);
end

[Bill Tubbs]

For others like me just discovering timetables, there are some good guides on the documentation page, covering all the basics including:

• Create Timetables and Convert Type
• Read and Write Files
• Select Part of Timetable
• Sort, Shift, and Synchronize
• Missing Values
• Stacked Plot
• Create Timetables
• Clean Timetable with Missing, Duplicate, or Nonuniform Times
• Resample and Aggregate Data in Timetable
• Combine Timetables and Synchronize Their Data
• Retime and Synchronize Timetable Variables Using Different Methods
• Select Times in Timetable
• Using Row Labels in Table and Timetable Operations
• Modify Units, Descriptions, and Table Variable Names
• Add Custom Properties to Tables and Timetables

Answer 2

[Bill Tubbs]

Here is my answer to illustrates the timetables solution proposed by @dpb in the comments above.

u = timetable(seconds(t),u1,u2,u3,u4);

This does work in the sense that you can now easily select the data for one column:

u(:,'u1')

or more than one column:

u(:,{'u1','u3'})

Timetables with one element are not supported by the plot function but you can use the stackedplot method:

stackedplot(u(:,'u1'))

Alternatively, you can get the desired column as an array like this:

plot(u.u1)

[Bill Tubbs]

I just noticed there is also a new function ts2timetable in version 2021a to convert timeseries objects to a timetable which could be useful—e.g. for converting outputs of Simulink 'to workspace' blocks to a timetable.

[dpb]

"...easily select the data for one column:"

u(:,'u1')

NB: that for the one column the "dot" syntax is even easier--

u.u1

while for mutiple variables one has to specify both a row and variable subscripting expression. I've thought it should be possible for the colon to be implied for all rows and just give a list of variable names, but I suppose it complicates the parser enough TMW hasn't added the feature.

NB: that addressing with parentheses "()" returns another table while using curly braces "{}" returns an array, similar to dereferencing a cell array to get the underlying content.

[Bill Tubbs]

The difference between u(:,'u1') and u.u1 is that the former returns a timetable object whereas the latter returns an array of the data. The reason I want the former is so that I can easily make plots with automatically-labelled axes and titles.

[dpb]

" make plots with automatically-labelled axes and titles."

Oh. I don't know that I've ever utilized the feature "in anger" (as an old former colleague/Scottish power engineer was wont to say of beginning a test)...