Repository navigation
GAM parameter names are non-standard #2857
Description
The trained Generalized Additive Models (aka GAMs) methods for obtaining trained model parameters have non-standard names and various inconsistencies.
These are the properties:
var shapeFunctionsBins = gamModel.GetBinUpperBounds();
var shapeFunctionsValues = gamModel.GetBinEffects();
var numShapeFunctions = gamModel.NumShapeFunctions;
var intercept = gamModel.Intercept;There are a few issues here:
- Nonstandard names
I proposeBinUpperBoundsandBinEffectsbe renamedShapeFunctionsBinsandShapeFunctionsBinValues. - Inconsistent names
Sometimes we sayShapeFunctions, sometimes we done. Same solution proposed as No. 1. - Inconsistency in
Getmethods vs. properties
e.g.Interceptvs.GetBinUpperBounds.
I propose we move all the methods to be properties.
e.g.model.ShapeFunctionsBins. - Names leak implementation
BinUpperBoundsvs.Bins.
Question: should we just sayBinsrather thanBinUpperBoundsand say that they are the upper bounds for the bin in the documentation?BinUpperBoundsis very verbose, and speaks to how we bin.
Activity
Ivanidzo4ka commentedon Mar 5, 2019
#2753 You can tell this to @wschin since he is doing scrubbing of fast tree learners.
rogancarr commentedon Mar 5, 2019
@wschin Re-assigned this to you!
wschin commentedon Mar 5, 2019
They look a bit long and the name
GamModelParameters.ShapeFunctionBinsdoesn't look more informative thanGamModelParameters.BinUpperBoundsbecauseShapeFunctionis rarely used elsewhere.It's consistent. For jagged 2-D array, we already return
IReadOnlyListover one of its slice in FastTree's APIs.I am hesitating to have
while the following looks clearer.
Cleaning of GAM family is also a part of #2753.
Explode column types and generate converters (dotnet#2857)
Move DataFrame to machinelearning (#5641)
Merge from main repository (#1)
Actually merge from main (#2)
Update editor config and format samples (#5893)