| The bias-variance decomposition |
|
|
|
|
| model’s
estimate |
|||
| for
test case n |
|||
| when
trained on |
|||
| dataset D |
|||
|
|
|
|
| average |
|||
| target |
|||
| value
for |
|||
| test
case n |
|||
|
|
|
| The
“bias” term is the squared error of the |
||
| average,
over all training datasets, of the |
||
| estimates. |
||
|
|
|
|
| angle
brackets are |
|||
| physics
notation |
|||
| for
expectation |
|||
| over D |
|||
| see Bishop page 149 for a
derivation using a different notation |