search

MSD-IRIMAS / Simple-KAN-4-Time-Series

A simple feature-based time series classifier using Kolmogorov–Arnold Networks
GNU General Public License v3.0
80 stars 12 forks source link

hydra configuration #3

Open dgiunchi opened 1 month ago

dgiunchi commented 1 month ago

I have some problem on the configuration of Hydra (totally new on that). From the default, in which way should I configure to make a run on one UCR dataset that I downloaded? What the width should be? Thank you!

hadifawaz1999 commented 1 month ago

Thank you for using our work ! did you download all of the ucr archive and mount the archive directory into the docker container as detailed in the Readme? The dataset is loaded using this function so you can see the root path is /home/myuser/ucr_archive/ which is inside the docker container, by simply having the dataset folders inside this directory you can change the dataset variable in the hydra configuration file to choose the dataset. The archive directory will contain a folder per dataset as such: /home/myuser/ucr_archive/ArrowHead/ /home/myuser/ucr_archive/Coffee etc.

For the width parameter, when set to Null, the KAN model will simply a one layer network that takes as input the time series Catch22 features and outputs C neurons where C is the number of classes in the dataset. However if you set width to be a list of integers, each integer represent the number of neurons in the hidden layer. For instance if you set width=[2,3] and the input dimension is 22 (because of Catch22) and the dataset contains 5 classes, then the total architecture will look like this:

input layer dimension 22, hidden layer with 2 neurons, hidden layer with 3 neurons, output layer with 5 neurons