Structure of Strings

Hi!

I want to specify possible parameters for the naive bayes classifier. For this reason, I have created a structure with a field for each parameter and a matrix with possible values.

bayesPosParams.distribution = ['normal'; 'kernel'];    
bayesPosParams.prior = ['empirical'; 'uniform'];       
bayesPosParams.KSWidth = 'scalar';                   
bayesPosParams.KSSupport = ['unbounded'; 'positive'];

Unfortunately that does not work: ??? Error using ==> vertcat CAT arguments dimensions are not consistent.

All Strings need to have the same length. What would be the best possibility to specify the parameters nice and consistently?

Thanks in advance, Jay

Answer 1

[Andrei Bobrov]

bayesPosParams.distribution = {'normal'; 'kernel'};    
bayesPosParams.prior = {'empirical'; 'uniform'};       
bayesPosParams.KSWidth = 'scalar';                   
bayesPosParams.KSSupport = {'unbounded'; 'positive'};

Answer 2

[Jay]

Hi Andrei!

Then every field would be a cell array, correct? What's the differences between this approach and using, my just discovered method, bayesPosParams.distribution = char('normal', 'kernel')?

I will want to operate on bayesPosParams with for-loops.

Thanks!

[Jan]

Using CHAR creates a rectangular matrix by padding shorter strings with spaces. Then comparing a row of the matrix with STRCMP is harder, because the trailing spaces have to be considered. Therefore the cell string is more comfortable and more efficient:

CHAR: unblank(bayesPosParams.distribution(i, :))

CELLSTRING: bayesPosParams.distribution{i}

[Andrei Bobrov]

Thanks Jan, I full agree with Jan

[Jay]

hi, thanks a lot!

If I'd do:

for dis = bayesPosParams.distribution

then dis would become 1x1 cellarrays ( {normal} and {kernel} ) right?

How can I get dis to get the two chararrays 'normal' and 'kernel'?

Thanks again,

Jay

[Andrei Bobrov]

no, dis would become 2x1 cellarrays

>> dis = bayesPosParams.distribution

dis =

'normal'

'kernel'

>> dis{1} % call char array

ans =

normal

>> dis{2} % call char array

ans =

kernel

[Jay]

right, i want to test all possible combinations of parameters.

only way i could think of is to do this with four nested for-loops. Probably very slow, but don't know how to do that else.

So I would for example want to iterate over all chararray-values of bayesPosParams.distribution, namely 'normal' and 'kernel'.

How do I do that?

Something like:

for dis = bayesPosParams.distribution{}

....

but this does not seem right either...

[Andrei Bobrov]

dis - cell array, and may be any operation with 'dis', as with cellarrays