Best practices for using isBlank, isDefined, ifBlank, and ifDefined

Using the right functions has a real impact on model performance, sparsity, and database computation efficiency.
Here’s a practical breakdown to help you decide the best formula for your use case — based on recent technical discussions with Pigment Engineers.

🚀 TL;DR

🔍 Key Concepts and Recommendations

ifBlank(A,B)

• Behavior: Acts like A if A is defined, otherwise returns B.
   
• Performance: Efficient; avoids densification unless B is dense.
   
• Best Practice:
  ✅ Prefer ifBlank over if(isBlank(A), B, A) for clarity and efficiency.
• Important note: as it looks very similar to isBlank and isNotBlank it is important to make clear that this function will not densify (unless metric B is dense).
   

isBlank and isNotBlank

• Behavior: Always outputs True or False, resulting in a dense metric.
   
• Impact: Densification can negatively affect model performance.
   
• Best Practice: ⚠️ Use cautiously, only use when you need False as an outcome.
   

isDefined

• Behavior: Outputs True if defined, otherwise Blank.
   
• Impact: Preserves sparsity, keeping models lighter and faster.
   
• Best Practice:  ✅ Favor isDefined over isBlank for better sparsity management.
   

ifDefined

• Behavior: Shortcut equivalent to if(isDefined(...)).
   
• Impact: No impact on database computation; purely a formula-writing shortcut.
   
• Best Practice:  ✍️ Use for cleaner model formulas when checking for defined values.
   

Why was isNotDefined not created?

• Introducing isNotDefined led to ambiguous cases mixing False and Blank.
   
• Workaround:
  Use if(isDefined(X), blank, true) if you need this behavior.
   

🎯 Final Thoughts

Even small formula decisions — like choosing ifBlank instead of if(isBlank()) — can make a big difference at scale.

By keeping models sparse and simple, we can build faster, cleaner, and more maintainable models.