Projects

The University of Chicago

GStore customer revenue prediction

  • Conducted exploratory data analysis, feature engineering and made inference along the way.

  • Predicted the natural log of the sum of all transactions per user using light gbm and xgboost and achieved 1.42 RMSE on test data

Exploration of External Memory in Variational Auto-Encoder

  • Proposed a VAE based model that has an memory matrix as in VQ-VAE, but with soft and differentiable extraction operation and achieved comparable results with faster convergence rate.

  • Implemented the new model with different frameworks(Pytorch and Tensorflow) on local machine as well as Google Colab and explored system related problem

Multiple testing and inference problems for Capital Bikeshare data

  • Participated in a group of four to implement multiple testing and inference problems for Capital Bikeshare data

  • Preprocessed bike-share dataset by data cleaning, continuous variable transformation and data structure reconstruction

  • Compare Spatial Analysis and Non-Spatial Analysis using BH procedure and group BH procedure

  • Do inference on Bicycle Demand using Lasso Regression, Forward Stepwise Regression and Least Angle Regression

A study on the global warming trend: evidence from Seasonal Central England Temperature data

  • Estimated the trend in seasonal temperature data using Polynomial and Periodic Regression Smoothers, and non-parametric smoothing methods, such as nearest neighbor, and splines

  • Designed and implemented an ARIMA model and performed out-of-sample test, compared the actual data and confidence band of prediction, and concluded that the warming trend does not slow down, but even become faster

Zhejiang Unviersity

Predicting price of stocks using improved KPCA and SVM

  • Conducted predicting algorithm using support vector machine(SVM) combined with PCA, KPCA, Isomap feature extraction

  • Proposed a new method to apply group lasso before KPCA and achieved comparable NMSE with 20% less time.

Research on the Implementation of Shanghai-Hong Kong Stock Connect to China Stock Market

  • Applied ARIMA model to simulate and forecast Shanghai Composite index before and after the official opening of Shanghai-Hong Kong stock connect program

  • Established models to make prediction and theoretical analysis and summarized the conclusion that the implementation of the Shanghai-Hong Kong stock connect program has certain positive influence on Chinese stock market due to the ascension of investor confidence and increase of the stock market capital inflows

Statistics Course Programs

  • Multivariate Statistical Analysis: Implemented factor analysis model to make prediction of Most Valuable Player from year 2011 to 2015 in NBA based on the player’s abilities involving offensive capability, credits and appearances

  • Regression Analysis: Selected appropriate data and established the appropriate linear regression model through variable selecting, modeling test, etc. for pork price modeling