Weijun's Blog

Weijun's Blog

Life is a circle.

16 Jun 2022

GSoC Part 2 - Construct Regression functions in queries with GROUP BY

Overview

In this part, I finished the design of regression function, and implemented most part of them. Basically, the implementation of window can be completed at the same, but the tests still need to be verified.

You can see my pr here

Design

The core part of these functions is how to accumulate the value instead of calculating at the end of process. I adopted the Welford’s online algorithm and implement it in Regr_result class. Then all of Item_sum_regr_XXX will use it to get the value by invoking individual bool val_real(). The following UML figure shows the architecture.

classDiagram Item_sum_double <|-- Item_sum_regr_sxx Regr_transition <--o Item_sum_regr_sxx Item_sum_regr_sxx <|-- Item_sum_regr_sxy Item_sum_regr_sxx <|-- Item_sum_regr_syy Item_sum_regr_sxx <|-- Item_sum_regr_avgx Item_sum_regr_sxx <|-- Item_sum_regr_avgy Item_sum_regr_sxx <|-- Item_sum_regr_slope Item_sum_regr_sxx <|-- Item_sum_regr_r2 Item_sum_regr_sxx <|-- Item_sum_regr_intercept Item_sum_regr_sxx: Regr_transition transivalues Item_sum_regr_sxx: bool add() Item_sum_regr_sxx: void reset_field() Item_sum_regr_sxx: void update_field() Regr_transition: double m_x Regr_transition: double m_y Regr_transition: double sx Regr_transition: double sy Regr_transition: double sxx Regr_transition: double syy Regr_transition: double sxy Regr_transition: ulonglong N Regr_transition: void recurrence_next(double nr_y, double nr_x) Item_sum_regr_sxy: double val_real() Item_sum_regr_syy: double val_real() Item_sum_regr_avgx: double val_real() Item_sum_regr_avgy: double val_real() Item_sum_regr_slope: double val_real() Item_sum_regr_r2: double val_real() Item_sum_regr_intercept: double val_real()

Implementation

You could see my code here

It is worth noting that the aggregation function with GROUP BY used the Item_regr_XXX_field to store the value. While the aggregation without GROUP BY used the Aggregator class to iterate each row.

comments powered by Disqus