link

When training ML model need Orthogonalization in order to Determine what to tune, and observe the effect it has.

Each button does one thing.

Example: Car with Accelerator and angle of steering wheel.

Assumptions (controls for tuning):

  • model works well with cost functions
  • work on training set
  • works on test set of data
    • Try bigger training set.
  • works well in real life.
    • Change training set
    • Change cost function.

Avoid early stopping as effects network size, and training set size.

Related links: