how does the lda model inference work on new document

[Rui Guo]

I would like to know the Gensim LDA model inference strategy. 

I also implemented my own methods to calculate each topic score:

Topic score:

For each topic

     For each document

           word_count = count the number of each word in the document

           topic_score += word_count * word_weight_in_that_topic

Then I use the softmax on the topic scores calculated to make it a probability distribution.

I see the topic score method above has a big different result from the LDA model inference by Gensim. Which topic score method should I use?