This week, me and my roommate participate in the kaggle competition. This one is proposed by otto group and it is to predict the distribution of class_i given an item with 93 features. The visualization can be found here.

We start from using stacked auto-encoder and multilayer perceptron but the cross validation only reach around 80% correctness.

Today,we adapt Gradient Boosting Machine( one of the ensemble method), I hope this will squeeze a few percent of the correctness. It’s still running. I’ll keep posted.