Data structure

[RobertBiehl]

Hi Been,

I liked your presentation a lot and would like to play around with the code on different data sets.
Can you give me a quick hint how to construct the data needed for your program?

Thanks,

Rob

[rjvak7]

Hi Robert, the repository contains a representative dataset (usps) that we used to run quantitative performance on. You can start at run_digits.py to see, it should run out of the box.

[RobertBiehl]

Yes, I did that, and I also got the image output working. 😊 My question was more targeted towards using other datasets. I presume it takes several steps to preprocess images and create the dataset file to make it work.

…

-Rob

On 8 Dec 2016, at 14:19, rjvak7 ***@***.***> wrote: Hi Robert, the repository contains a representative dataset (usps) that we used to run quantitative performance on. You can start at run_digits.py to see, it should run out of the box. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

[rjvak7]

Ah ok, great! the usps dataset is in libsvm format with <feature_1:value_1> .... <feature_n:value_n>. You should be able to directly use the code as long as you can get your data to this format. Or, alternatively, you could add your own functions to load data (see data.py) and call prototype/criticism functions from mmd.py . Both methods are equivalent in terms of effort involved and output.