HOW-TO: Publishing your data
This page should tell you everything you need to know in order to publish your data set on UCSD-MKL.
- The first step is to create an account.
- Once your account is active, the next step is to add the related paper to the bibliography section. This will allow you to connect your data set to the original publication, and any subsequent relevant work.
- Next, go to Create content and create a Data Set entry. You will be prompted for some basic information about your data set (e.g., name, description, number of points). Please upload your data as a single archive (.zip or .tgz) which includes kernels and/or features (see below), labels, and any other relevant information.
- Once the data set has been created, we will review it before publishing it in the repository.
- Please store numerical data in either .csv or Matlab format.
- Since the data used to generate kernels may be much larger than the kernels themselves, we do not require all data sets to include raw features. If possible, please provide a link to the original data set.
- Similarly, some MKL algorithms operate directly on features to generate a large number of kernels, which would be too large for us to store. For these reasons, we do not require you to upload both features and kernels, but if you must omit one, please indicate why.
- When uploading kernels, please describe what kernel function is being used for each kernel, and any pre- or post-processing (e.g., centering and normalization).