Elegant way to batch extract all as single data from Data Pool

I am Matlab beginner, which to know if there's a quick way of indexing/slicing data into individual variable.

dataPool=magic(5)

dataPool = 5×5

17 24 1 8 15 23 5 7 14 16 4 6 13 20 22 10 12 19 21 3 11 18 25 2 9

Naive method is

p=dataPool(1,1);
q=dataPool(1,2);
r=dataPool(1,3);
s=dataPool(1,4);
t=dataPool(1,5);

Can I do a batch processing on this? For instance:

[p,q,r]=dataPool(1,:)

Insufficient number of outputs from right hand side of equal sign to satisfy assignment.

I pretty sure there must be some methods published in Mathworks, but it is hard to describe this problem as search string/keyword to look for help. Let me know if you know one webpage that talk about this. Your help is very much appreciated.

Thank you.

More information:

I am developing an algorithm that produces input matrix which composed of m*n (m runs* n parameter). Matrix example:

___________________________________________________

| Parameter1 Parameter2 Parameter3

____________________________________________________

Run#1 | 3 grams 45 degreeCelcius 3 bar Pressure

Run#2 | 4 grams 30 degreeCelcius 4 bar Pressure

These matrix will be extracted row-by-row to run a chemistry experiment, and the experiment program reads by variables' name (because these variables will be processed individually for instrument controller to start functioning), so each parameter must have a unique tag/variable_name. I am looking for a way, that helps to reduce my line of codes (i.e. Parameter1=dataPool(:,1);Parameter2=dataPool(:,2);) ,because i have long list of parameters

The convenient method is definitely the simplest way out. Just out of curiousity, if anything else works.

[Stephen23]

" if there's a quick way of indexing/slicing data into individual variable."

Indexing.

"i.e. Parameter1=dataPool(:,1);Parameter2=dataPool(:,2);)"

Numbering variable names (or more generally, forcing meta-data into variable names) is a sign that you are doing something wrong.

"so each parameter must have a unique tag/variable_name."

Variable names should not matter. Much better data design would for example store all of the meta-data and data in one table, making it trivial to access (and allowing you to really simplify and generalize your code).

[dpb]

@Stephen23's comment +inf

Reinforces what I've been preaching...

Answer 1

[Walter Roberson]

Expand = @(c)c{:};

[p, q, r, s, t] = Expand(num2cell(dataPool(1,:)));

That is, no built-in way is provided to split numeric vectors or arrays into different outputs. A built-in way is provided to split cell array into outputs, and a built-in way is provided to convert a numeric array to cell, but the allowed syntax does not permit combining those two without ugly hacks or using a helper function.

[Cha Yong Jong]

This is exactly what I am looking for! Thanks

[dpb]

Almost certainly NOT the way to proceed in writing your code effectively, however...

[Cha Yong Jong]

@dpb @Stephen23 Thank you for the professional guidances. It's truly sad to have this kind of indexing strategy which slows down the computation. I think MATLAB community's masters have been coaching a lot towards full utilization of MATLAB power. I believe developer will be upset to see my post.

However, I was just trying to establish a communication between two MATLAB functions, which the first function produces a MATLAB matrix (i.e. DataPool), and the second function takes in by variable name. The easiest way out is to use @Walter Roberson 's answer. (ps, I can't put my codes up since they are confidential for research)

First function

[DataPool]=Optimizer(a,b,c,d)

Second function

[Controller Input]=RunExperiment(DataPool_Parameter_1,DataPool_Parameter_2,DataPool_Parameter_3)

Quick way to my problem: Walter's

_____________________________________________________________________________________

However, it is worth to explore into modification in large scale to speed up entire system from function 1 to communication to function 2.

@Stephen23 gave an idea about converting the matrix into table. If I understand that correctly. Say, if I've modify the second function to takes in table as variable. How do I properly index for the value that I am looking for. Is it?

Arrhenius(Table.(Parameter1).1)

Other than the disadvantage of taking up some spaces (kill memory) through occupancy of Value1 in Workspace. Does this make a difference in term of computational speed as the one Walter suggested

Value1=Expand(num2cell(dataPool(1,:)));

Arrhenius(Value1);

Please be noted in the second function, the matrix's values are used all over the place. i.e. Temperature is used in Arrhenius equation, Pressure is used in gas Law, to further compute parameter that the controller can understand. Good day

[Stephen23]

"The easiest way out is to use..."

Basic indexing:

tmp = Optimizer(a,b,c,d);
[..] = RunExperiment(tmp(1),tmp(2),tmp(3),..,tmp(N));

[dpb]

The application of the anonymous function to make the conversion for the purpose is a needed/useful instance that is "the exception that proves the rule" to use the old saw.

The advice against coding with multiple sequentially-named variables or with variables containing metadata in their names is still good for the general case we see when asked the question as your initial posting did without the specific application -- the beginner approach is very often to do just that and then for every one of those variables end up copying the same code line over and over where instead a single vectorized expression could have been written elegantly. Or, the immediate next question raised is how to create those sequentially-named variables dynamically. It is in that context that the advice against doing so was given.

MATLAB is good about not making copies of argument data in functions if the data are passed in as an array and then local variables are created from elements of the array; it will use references to the array addresses instead, so that an alternative I would suggest if the function RunExperiment() is your function and not one you can't modify would be to change its syntax to pass the array and move the code to create the individual variables from the array inside it, out of the higher-level user code.

Coincidentally, the same question was asked more directly to the objective just yesterday <here>.

[dpb]

Stylistic note -- if were really going to go this route, NB that you could make the top level user code a little simpler by defining dataPool as a cell array to begin with -- then could expend with the helper function entirely as all it really gains you is dereferencing the cell array that you've had to create manually in the argument...

dataPool=num2cell(magic(5));        % the array as cell array instead
 [p, q, r, s, t]=dataPool{1,:}      % split to variables instead...

p = 17

q = 24

r = 1

s = 8

t = 15

As an aside, I've always thought this is a case deal should handle (maybe with a user option) differently; if the number of elements of the RHS matches the number of targets, assign 1:1 instead of duplicating as it does...

Answer 2

[dpb]

The real answer in MATLAB is "Don't Do That!" -- making variables out of array elements generally is not the way to write efficient MATLAB code and utilize its primary reason for being -- namely, the vectorized functionality.

This is a hard concept for those who come to MATLAB from other procedural languages, granted, but is key to making effective use of MATLAB.

Use array indexing instead...

Granted, occasionally it is convenient to address array segments as variables for convenience such as in your above example

x=dataPool(:,1);
y=dataPool(:,2:end);
plot(x,y)

lets you write a shorthand for the plot() function or other operations on the independent variable and its corollary dependent ones. Almost all MATLAB builtin function operate by column by default so the above plots the four columns against the first. Taking advantage of this is the key to effective and efficient MATLAB code both from the coding side as well as then performance when running.

What would be the end purpose of creating all these named variables that prompted the Q? Given a use case, we can probably illustrate "the MATLAB way" instead...

ADDENDUM

NOTA BENE: If x,y above are not manipulated numerically, then behind the scenes MATLAB will NOT create physical copies of the data but only references to the original. Hence, there is not a penalty for the coding convenience in this case.

Where the problem arises is that one you have created your above named variables, then you have no recourse going forward to use them but to use those specific names -- any generality is lost except by the arcane and performance-killing use of the not-to-be-mentioned-in-polite-company eval. And, you definitely do NOT want to even think about going down that route.

[Cha Yong Jong]

That's a insightful response. Thank you. Vectorized functionality is definitely an powerful tool.

Pardon my question that short of information. I am developing an algorithm that produces input matrix which composed of m*n (m runs* n parameter). Matrix example:

___________________________________________________

| Parameter1 Parameter2 Parameter3

____________________________________________________

Run#1 | 3 grams 45 degreeCelcius 3 bar Pressure

Run#2 | 4 grams 30 degreeCelcius 4 bar Pressure

These matrix will be extracted row-by-row to run a chemistry experiment, and the experiment program reads by variables' name (because these variables will be processed individually for instrument controller to start functioning), so each parameter must have a unique tag/variable_name. I am looking for a way, that helps to reduce my line of codes (i.e. Parameter1=dataPool(:,1);Parameter2=dataPool(:,2);) ,because i have long list of parameters

The convenient method is definitely the simplest way out. Just out of curiousity, if anything else works.

[dpb]

Would need the details of just how you interact with the controller, but I'd build the database as a MATLAB table instead of arrays; each column can hold the various parameter value for each set of test conditions, the first column would then be variable "Run" and the value 1:N (or, that can be implied if it is sequential but you may want to randomize run sequence for control of experimental variance).

Then, to run the sequence of experiments, simply iterate over the height() of the table and return/submit each data set in turn.

PS. Do NOT store metadata like units with the data values itself; the table actually has a units property you can associate to each; treat data as data and never the twain mix.

[dpb]

"... developing an algorithm that produces input matrix ... to run a chemistry experiment,"

If you have the Statistics Toolbox, it contains a pretty good selection of experimental design tools D(esign)O(f)E(xperiments) or DOE that might possibly be of interest/help in setting up the design matrix.

Includes full/fractional factorials, RMS (response surface model-specific), various D-Optimal designs, etc., etc., ...

[Cha Yong Jong]

Right. I am conducting something like DOE, but a modified-DOE to suit the chemistry science limitation. Thank you for your inputs. They are helpful