FunctionRacerProblemsRace

Drop the Weak Ones

HardstatsmodelsPython

df has a column y and several predictor columns. Trim the model by backward elimination:

1. Fit y on all predictors still in the model (with an intercept).
2. If the largest p-value among the predictors is above 0.05, remove that predictor and go back to step 1.
3. Otherwise stop.

Return the names of the predictors left, in alphabetical order (possibly none).

Example 1

df {"a":[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15],"b":[5,3,8,1,9,2,7,4,6,0,3,8,5,1,9],"y":[3.2,4.9,7.1,9.2,10.8,13.1,15.2,16.8,19.1,21.0,23.2,24.9,27.1,28.8,31.2]}
returns ["a"]

y climbs by about 2 per unit of a. Column b is noise: its p-value is far above 0.05, so it goes, and a stays.

Example 2

df {"a":[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15],"b":[5,3,8,1,9,2,7,4,6,0,3,8,5,1,9],"y":[6.2,4.1,5.9,7.2,4.8,6.1,5.2,6.8,4.1,5.0,7.2,4.9,6.1,5.8,5.2]}
returns []

Neither column explains y. b goes first, then a.

Constraints

Race itor take your time

Write it in your browser, against the clock. The model solution, stepped through line by line, comes after.

More statsmodels problems