How to adopt to custom dataset? about gsn HOT 1 OPEN

Hello,

There are a few modifications that need to be done. I will guide you through them below.

First, create a folder named XXX, inside the root folder that you have selected (--root_folder argument set from the command line/main.py file), where XXX is the name of your dataset family (--dataset argument). In case you have dataset subcategories you can create folders named YYY inside root_folder/XXX/ (--dataset_name argument). If not, YYY can be left empty (''). Place your raw data in root_folder/XXX/YYY.

Then, add the name of the new dataset family to the accepted list of names here (L211). Similarly, add a new case here (L46), such that the load function that you will create next (see below) is called when the corresponding --dataset and --dataset_name arguments are set in the main file.

Finally, you need to add a load function here that takes as input the path to your raw data and returns (1) a list of graphs (each element in the list contains all the information for a graph, e.g. node features, edge list etc. - see the examples in this file for more info) and (2) the output dimension (e.g. number of classes for classification tasks).

The rest should be taken care of by the existing code, as long as there is consistency in the variable names. To choose the substructures see here.

I hope this helps.

from gsn.