NLP: building a sentiment model

I just began learning NLP so at novice level. I want to create a sentiment model. The dataset is high dimensional with many TF-IDF columns and about 4 sentiment columns negative, positive , neutral and compound. I thought the compound sentiment values was the total sentiment to be used as my target, however on visual examination the other sentiment numbers don't seem to add up to the compound sentiment. The sentiment values are also very large or very small sometimes …

began building create dataset however idf languagetechnology negative nlp positive sentiment tf-idf thought total values visual