Best shap for data to be in to work on.

Hi,

I have a number of years of time series data of the format;

Date  Time  Open  High  Low  Close
734141  1544  1.6152  1.6152  1.6152  1.6152
734141  1545  1.6152  1.6153  1.6151  1.6152
734141  1546  1.6152  1.6152  1.6152  1.6152

This is stored in an Access DB that I am able to import change dates, times, manipulate the data ect..

I would like to do analysis on data per day, such as Max(time_a, time_b), HighLow(t_a, t_b), price_direction(after_t_c), MovAv, interval_diff.

What I am having trouble with is the shape of my initial stored data in matlab. Should I store it as a single array of time series data, a double array or would a struc for each day be better. I eventually would like to run my complete data set of day data through functions I build, testing my models and ideas, creating flags when criteria are met.

I know this is a slightly unusual question as you all usually deal with code/math issues that can be solved by debugging or writing/suggesting code. If your unable to help would some one be able to point me in the direction of a matlab forum/group that might be able to advise. I am unemployed and not in a class or college so I don't have colleagues or profs to ask or discuss with.

Many thanks.

AD

Answer 1

[Fangjun Jiang]

It looks to be a good use case for dataset(). See doc dataset.

Also see this blog. http://blogs.mathworks.com/loren/2009/05/20/from-struct-to-dataset/

[Scragmore]

Thanks, I have taken a look unfortunately I seem to have gone backwards. I was able to create a simple data set;

d1 = dataset({datetest3(:,1), 'year'}, {datetest3(:,2), 'time'},{datetest3(:,3), 'open' },{datetest3(:,4), 'high' }, {datetest3(:,5), 'low' },{datetest3(:,6),'close' });

six columns of data with respective headings. But that is all I have been able to do so far. I have tried a simple query where I return time and open price for a specific day but get either an error because of my syntax or an empty dataset. I've tried ;

dd2 = d1.year('734260');

dd2 = d1.year('734260', {'time', 'open'});

dd2 = d1.year(734260, {'time', 'open'});

dd2 = d1('734260', {'time', 'open'});

dd2 = d1('year' == 734260, {'time', 'open'});

dd2 = d1('year' == '734260', {'time', 'open'});

but none of them seem to work. I am not sure this is what I was after as it just seems considerably more complicated. What am I missing.

http://www.mathworks.co.uk/help/toolbox/stats/dataset.html

http://blogs.mathworks.com/loren/2008/07/23/using-matlab-to-grade/

http://www.mathworks.co.uk/help/toolbox/stats/dataset.datasetfun.html

http://stackoverflow.com/questions/5811805/matlab-dataset-question

http://en.literateprograms.org/MATLAB_Financial_Toolbox

Just a few of the pages I have tried to find a solution on.

[Fangjun Jiang]

I regard dataset() as an enhancement of structure. I have not used it very often so I have to learn it myself too. I believe it can be done. The nice part of it is that you can use those header text such as 'time','open','close' to reference your data thus make it much more readable.

You can use regular 2-D array to achieve the same if it's more comfortable or easier for you. You probably just need to use logical index for most of your query. For example.

index=data(:,1)==734141;

SelectedData=data(index,:);

This way, you just need to remember which column is Date, which column is Time, etc.

[Scragmore]

Hi,

Thanks for your time and suggesting this. Unfortunately I am still unable to work out how to query the dataset.

I shall continue to look for some help on how I should construct my data.

Thanks and regards,

AD