I'm working on a machine learning project where I'm trying to predict the revenue of a movie.
My dataset contains mixed data types. There are numerical features (rating, number of votes, release year,...), categorical features (genres, studio, is the movie for mature audiance,...) but also embeddings that consists in large feature vectors (post embeddings and movie description embeddings).
My problem is with the last data type. I'm wondering how should I handle these embeddings ?
I've made some pre-processing (cleaning, one-hot encoding, label encoding,...) but I still have these embeddings. So basically, now I would like to do some feature selection and model selection but for example, let's say I would like to do a filter method. For a linear model, I can use a correlation matrix but I cannot compute it since the variable
txt_embeddings are not numericals but 1D vector. Same if I want to use mutual information for non-linear models.