← All workshopsRead Loss Functions

Neural Networks

Measure the prediction errors

Which miss dominates the second batch?

Botie's predictions include misses in both directions. Averaging signed errors could conceal those mistakes, so the score needs a different treatment.

BOTIE

The predictions are 2, 6 and 1 against targets of 3. Which observation contributes most after squaring?
The inputs we are comparing

Predictions

Example 1

  1. 01
  2. 12
  3. 25

Example 2

  1. 02
  2. 16
  3. 21

Answers

Example 1

  1. 03
  2. 12
  3. 23

Example 2

  1. 03
  2. 13
  3. 23
Result for example 1
2.6667
Choose an explanation to test
Three exact predictions and one large miss

This check uses the pieces from “Prevent overestimates from cancelling underestimates”.

BOTIE

Three predictions agree with their targets. The fourth misses by much more. Return mean squared error across all four observations, including the ones whose error is zero.
Predictions
  1. 04
  2. 14
  3. 24
  4. 34
Answers
  1. 04
  2. 14
  3. 24
  4. 312

Use a number, text, or an array matching the requested result. For a matrix, put each row in brackets. Saving your prediction runs your current construction on these inputs and checks the earlier examples too.

Build a construction in the workspace above first. This case will test those pieces.

Why would summing squared errors make comparisons between different batch sizes awkward?

The reduction and its denominator are part of the loss definition.

This is a small construction problem. The lesson explains the assumptions and the parts of the full method that this example leaves out.

Work through the Python challenges →
Continue the courseBackpropagation →
← The Shape Guarantee