Apoc.math.regr calculations

raghad.baiad · June 11, 2021, 9:19am

Hello Neo4j knowledgable community,

I have a question regarding one of the apoc math functions, namely apoc.math.regr I have been using it to compute the linear regression estimators (slope, intercept) and the goodness-of-fit measure (R-squared). However, I noticed there is a huge difference between what is resultant out of this function and what is computed manually and using other tools adopting the least square approach. Therefore, I would appreciate your advice in figuring out/understanding what is used in the computation using the apoc regr function

michael.hunger · June 24, 2021, 12:32am

I think we just use apache.commons.math there, feel free to have a look at the source

github.com

neo4j-contrib/neo4j-apoc-procedures/blob/4.2/core/src/main/java/apoc/math/Regression.java#L34


      
              public double slope;
          
          
    public Output(double r2, double avgX, double avgY, double slope) {
                  this.r2 = r2;
                  this.avgX = avgX;
                  this.avgY = avgY;
                  this.slope = slope;
              }
          }
          
          
@Procedure(name = "apoc.math.regr", mode = Mode.READ)
          @Description("apoc.math.regr(label, propertyY, propertyX) - It calculates the coefficient " +
                  "of determination (R-squared) for the values of propertyY and propertyX in the " +
                  "provided label")
          public Stream<Output> regr(@Name("label") String label,
                                     @Name("propertyY") String y, @Name("propertyX") String x) {
          
          
    SimpleRegression regr = new SimpleRegression(false);
              double regrAvgX = 0;
              double regrAvgY = 0;
              int count = 0;

and the tests.