Octo+

Formula extractor from input/output specs (using industry-grade tentacles).

Challenge: to create a proof of concept of a system that proposes potential formulas given the fields with inputs/outputs

In this project, we are focusing on machine learning approach to fit the independent variables (base, rate, days difference, etc.) to predict dependent variable (fee).

The main idea is that with polynomial regression (we use polynomial regression with 3 degree), it is feasible to predict coefficients with all possible variables combinations. Furthermore, we remove redundant coefficients in polynomial via Lasso Regression (which shows which coefficients are actually significant, and which should be removed).

Prototype is provided through web server.  

Demo video: https://vimeo.com/737355433