Sifan Guo

It was my first time to deal with 25+Gb dataset in csv files. I was responsible for the database retrieval and data cleaning as well as connect the csv files by certain IDs. As for the prediciton part, we relied on SPSS and used multi-linear regression because of its interpretation. The prediction result was not bad, and the test RME is acceptable. After that, I designed an interaction using C++ in Qt 5.7 which was very user-friendly.

This westwing transcript website provides an easy way for westwing fans to search for the transcript. Users could select the desired filter options on the left of this website and then search for the transcript.

Dimensionality reduction was applied to reduce the level of categorical features and the best number of components was determined. We implemented and compared light gradient boosting model (LGBM) and long short term memory (LSTM), and achieved a best validation root mean squared error of 0.389.

This is a binary classification prediction problem. Initially, I tried to extract image features on my own by downloading all the images and convert them into matrix format. But I do lack computational power, so I turn to Microsoft Computer Vision API which helps to digest the image data and return labels. And I use these labels as my features to predict if the question inside a image is "answerable". I get a 0.73 accuracy score.

In this project, I created a random number generator based on the running status. But it's worth noting that I didn't use the python package for md5 hash function. Instead, I wrote my own md5 hash function.