I was reading through the code and something caught my attention. In the implementation of KNN, the training_data features have been normalized and then in a matrix multiplication form, they are multiplied by the features of the eval_data. It looks like you are looking at something like Cosine Similarity but not exactly (cause you don't normalize the eval_data). Is there a reason why you did this? Or is it just a mistake?
Hi there.
I was reading through the code and something caught my attention. In the implementation of KNN, the training_data features have been normalized and then in a matrix multiplication form, they are multiplied by the features of the eval_data. It looks like you are looking at something like Cosine Similarity but not exactly (cause you don't normalize the eval_data). Is there a reason why you did this? Or is it just a mistake?