Multiclass logistic regression - most relevant features per class

Question

I want to use Logistic Regression for Sentiment Analysis on data with class 1.0 to 5.0.

1. Is it correct that I do not need LabelEncoder in this case because the t