Partial derivatives

Why it matters

Real functions have many inputs. Conversions depend on search spend and social spend and email volume; a model's loss depends on every one of its weights. A partial derivative asks the one-at-a-time question: if I nudge just this input, how does the output change? With one answer per input, you know which lever matters most right now, which is exactly what training needs to know about each weight.

What you'll be able to do

• Read functions of several inputs, like conversions from two ad channels
• Compute a partial derivative by holding the other inputs fixed
• Estimate partial derivatives numerically, one input at a time
• Use partial derivatives to decide where the next dollar should go

Before this lesson

• Derivatives

Next in this phase: Gradients