Learning Curve

A learning curve shows the validation and training Accuracy for varying sample sizes of training data.

It can be used to tell whether the model is currently Overfitting or Underfitting depending on the amount of data that was used to train the model.

![[CleanShot 2023-10-04 at 10.29.07@2x.png]]

The shape of the learning curve shows whether a given problem (target function) is realizable given the Hypothesis Space that was used for training.

• non-realizable → hypthesis space too small
• redundant → data redundancy, useless features, etc.
• realizable → hypothesis space has right size

![[CleanShot 2023-10-04 at 10.30.17@2x.png]]

Also see Validation Curve.