CF
Certifyforce
Einstein Discovery12% of the exam

Model variable settings, cardinality and bias analysis

Refining a model with Edit Model and Update Model: choosing variables by importance and correlation, avoiding overfitting, handling cardinality, bucketing and filtering values, and flagging sensitive variables with Analyze for Bias, all without touching the dataset.

Official exam objective
Given derived results and insights from the model, adjust data parameters and add/remove data or columns to improve the model.

~7 min read · 7 practice questions ·

What the exam expects

  • Edit a model's settings and create a new version with Update Model, knowing the dataset stays unchanged.
  • Decide which variables to keep from importance and correlation, and which overly detailed variables to drop.
  • Explain how Einstein Discovery treats low-cardinality numbers and high-cardinality text, and when to enable high cardinality.
  • Configure a variable: buckets, Include Only filtering or grouping into Other, and transforms.
  • Turn on bias detection with Analyze for Bias and act on the disparate impact result.

Key terms: model version (a new copy of the model created each time you create or update it; earlier versions are kept); importance (how much the model uses a variable, including interactions); correlation (how strongly one variable is associated with the outcome); cardinality (the number of unique values in a variable); sensitive variable (a variable flagged with Analyze for Bias).

Key facts

Update Model
Detail
Creates a new version; earlier versions kept
Source
Track Model Versions (opens in a new tab)
Model settings and the dataset
Detail
Settings and transforms change the model only
Source
Select Data Options (opens in a new tab)
Variables per model
Detail
At least 2, no more than 49
Source
Edit Model Settings (opens in a new tab)
Low-cardinality number
Detail
Ten or fewer unique values becomes text
Source
About Models (opens in a new tab)
Values per variable
Detail
100; 200 with high cardinality
Source
Select Data Options (opens in a new tab)
High-cardinality variables
Detail
One per model at up to 200 values
Source
High Cardinality Alert (opens in a new tab)
Rare values in Include Only
Detail
Under 25 occurrences grouped as Other
Source
Edit Model Settings (opens in a new tab)
Default algorithm
Detail
GLM; Model Tournament runs all four; multiclass supports no GLM
Source
Select Data Options (opens in a new tab)Multiclass Use Cases (opens in a new tab)
Disparate impact threshold
Detail
Selection rate under 80% of the highest group
Source
Disparate Impact Alert (opens in a new tab)
Sources for this section (7)

Check yourself

  • Why does a partner value with a dozen rows appear as Other in Include Only?
  • When should you apply value filters relative to transformations?
  • Which tool merges text categories that differ only by spelling?
  • What does Model Tournament do?
  • Why does Help call 80% the disparate impact threshold?

The questions are free to read. Answers with their sources and self-grading are included with the pack.

Keep reading with the CRM Analytics and Einstein Discovery Consultant pack

  • Core concepts
  • Exam traps
  • Worked scenario

The pack includes:

  • All 26 lessons of the CRM Analytics and Einstein Discovery Consultant guide, with reading progress
  • Interactive decision walks and step-by-step sequence steppers
  • Practice exams purchase adds inline bank questions and topic practice (7 questions for this lesson)
  • An official source cited for every claim
  • An AI tutor that answers from the guide
  • Choose guide-only, practice-only, or save $4 with both
Already own it? Sign in

Practice for the CRM Analytics and Einstein Discovery Consultant exam