I used it in sign language recognition by considering the entire sequence as a single pattern
My understanding is that PCA compresses features into few components and that this may help improve model performance. What if stakeholders need to understand which features are most important? Can those PCA components be expanded back to their original feature forms to obtain feature importance scores?
This awesome, now I can train my model faster 💪💪
And LDA
This is 👍
What are your thoughts on UMAP for high dimensionality data sets?
you make it sound so simple! :) ... so where's the catch? ;)
Is feature scaling required before pca?
how do you deploy it after pca?
Is there any version for deep learning?
you lose transparency when you're using PCA just an FYI. Usually this is very hard to explain to stakeholders/customers and if they don't understand it, they won't like it
Doesn't this underfit the model?
What about mutual info regression from skleaen.feature selection ?? Any difference in efficiency????
It doesn't always work, if you have categorical data it won't eork
Sounds good, i have like 1700 features, but after reducing the dimensions, how would I infer the model? Like what is the input after reducing the dimensions?
@nimeshkumar9613