Low-dimensional Interpretable Kernels with Conic Discriminant Functions for Classification
We propose several simple feature maps that lead to a collection of
interpretable kernels with varying degrees of freedom. We make sure
that the increase in the dimension of input data with each proposed
feature map is extremely low, so that the resulting models can be
trained quickly, and the obtained results can easily be
interpreted. The details of this study is given in our
paper.
All our codes are implemented in Pyhton 3.7 and we use the following
packages:
We provide the following tutorials to demonstrate our implementation.
For the proposed feature maps, we refer to the pages
FeatureMaps and Kernels. We
also provide the same tutorials as two notebooks, notebook one and
notebook two, respectively.
To obtain a row of Table 1, we refer to the page Table
1 or to the notebook.
To obtain a row of Table 2 and Table 3, we refer to the page
Table_2_3 or to the
notebook.
To obtain Figure 3, we refer to the page Figure_3
or to the notebook.
To obtain Figure 5, we refer to the page Figure_5
or to the notebook.
We provide the following scripts to reproduce the numerical
experiments that we have reported in our paper.
In this tutorial, we explain how to
apply the proposed feature maps with logistic regression on
binary and multi-class classificiation problems. Notebook version of
this tutorial can be found here.
We also provide the following code snippets that reproduce Tables
2-3 and Figure 5 in our paper, but this time, with logistic regression.