a data-centric way to build predictive models
The ML problem
Supervised regression learning
Robot navigation example
How it works with stock data
Example at a fintech company
Price forecasting demo
QuantDesk
factors we are using now <= choices of these factors are from another genetic algorithm
<= roll back time, and we look over all this last three months and look forward one month, see how accurate all those predictions were
https://lucenaresearch.com/#register
https://quantdesk.lucenaresearch.com/#login
Backtesting
ML tool in use
orange line => historical value of our portfolio
blue => benchmark (S&P500 here)
Problems with regression
Problem we will focus on
Parametric regression
K nearest neighbor
How to predict
Kernel regression
Kernel regression is different from KNN, because it uses kernel to weight the contribution of each nearest point
Parametric vs non parametric
Yes, the cannon ball distance can be best estimated using a parametric model, as it follows a well-defined trajectory.
On the other hand, the behavior of honey bees can be hard to model mathematically. Therefore, a non-parametric approach would be more suitable.
Training and testing
typically: train on older data; test on newer data
look ahead bias occurs if training reversely
Learning APIs
Example for linear regression
Note: This is intended to be pseudo-code only, although some Python-specific syntax has been shown.
原文：https://www.cnblogs.com/ecoflex/p/10977432.html