Abstract This prototype finished during P&G Data Science Hackthon, Nov. 2016 in Cincinati, OH. This document briefly elaborates the design ideology for SKII onsite recommender system, leverage the demographics data, skin scan result and purchase history. Note. all the confidential data has been pre-processed. The Dataset The data used in this recommender system consists of the folling 3 parts: Demographics information (1 million rows * 10 columns), including gender, age, marital status etc. READ MORE

Demo This is my recent play-for-fun, a tiny trick developing a weatherman robot on SNS. E.g. This robot will update the HK local weather everyday at 20:00 EDT and post it on my SinaWeibo account. Checklist A SinaWeibo developer account to use its post API (register here. You will need either a personal webpage or an application under development for registration). A WorldWeatherOnline API to update weather. They provide free API to fetch the weather worldwide and pack it in popular formats like xml, json or csv. READ MORE

A Gaussian mixture model is a probabilistic model that assumes all the data points are generated from a mixture of a finite number of Gaussian distributions with unknown parameters. Instead of having an output as cluster no. that a certain sample belongs to, Gaussian mixture model can additionally give an probability of this kind of clustering. In practice, expectation-maximization (EM) algorithm is often used for fitting mixture-of-Gaussian models. For better understand the theory of GMM this link could be followed. READ MORE