Training vs. Testing

• Training set:
  • Used for inducing the function.
• Testing set:
  • Used for evaluating the quality of the induced function.
  • Does the function "generalize" to previously unseen inputs?

Inductive Bias

• What assumption is made in order to generalize?
• Example algorithm:
  • Training:
    • Create a hash table with one entry (a linked list) per label.
    • For each training example:
      • Add it to the list corresponding to its label
  • Deployment:
    • If the presented data is contained in a list, return its label.
    • Otherwise...