We are compressing high dimensional vectors to 2 dimensions using PCA, so that we can make 2D SVM plots. But it is difficult to compress to 3 dimensions with PCA.What is a good way to compress to 3 dimensions to make 3D SVM plots?
EDIT: Actually, we can compress to 3 dimensions, but it was more difficult to plot than 2 dimensions. So the question is about how to plot 3 compressed dimensions.
Dear Christopher,
Your query is very interesting, I suggest some of its clue to follow https://towardsdatascience.com/support-vector-machines-svm-clearly-explained-a-python-tutorial-for-classification-problems-29c539f3ad8
It depends on what kind of plot you want to achieve? I assume you want to plot the hyperplane decision boundary of the SVM? There are some tutorials on that:
https://community.plotly.com/t/how-to-plot-3-d-boundary-for-any-kernel-svm/20906/2
https://stackoverflow.com/questions/36232334/plotting-3d-decision-boundary-from-linear-svm
https://towardsdatascience.com/visualizing-support-vector-machine-decision-boundary-69e7591dacea
Maybe those can help you out. I personally like the last one, although its just two components it uses the y-axis to plot decision values and it helps to evaluate the SVM results in my opinion. Maybe thats something you are looking for.
maybe you will find what you looking for in this page :https://stackoverflow.com/questions/36232334/plotting-3d-decision-boundary-from-linear-svm
It looks like a lot of these examples are for SVM models with a small number of features. I have over 1,000 features for each vector, so I still would need to compress them into 3 features. I can compress them to into 2 features with PCA. In the third suggestion by Kai Heinrich ( https://towardsdatascience.com/visualizing-support-vector-machine-decision-boundary-69e7591dacea ) they use the decision score as a third dimension in addition to PC1 and PC2. That might be a good option in my case. But the decision score for the compressed 2 dimensional model is different than that of the high dimensional model. But is it ok to use the more accurate high dimensional decision score with the compressed 2 dimensions?
- Badges
- Science topic