← All workshopsRead Do We Need a Learned Model Yet?

Define the ML Problem

Cost the inspection policy

Compare the candidate with the heat rule

Botie can compare a heat rule with a candidate model only after charging both for inspections and missed faults on the same pump records.

BOTIE

The candidate inspects both known faults and one healthy pump. What would justify preferring it on these supplied records?
The inputs we are comparing

Confirmed faults, 1 = faulty

Same in both cases

  1. 01
  2. 10
  3. 21
  4. 30

Inspection decisions

Heat rule

  1. 01
  2. 11
  3. 20
  4. 30

Candidate model

  1. 01
  2. 10
  3. 21
  4. 31

Cost per inspection

Same in both cases

3

Cost per missed fault

Same in both cases

20
Result for heat rule
26
Choose an explanation to test
Inspecting everything avoids misses but still costs money

This check uses the pieces from “Price each consequence of the inspection policy”.

BOTIE

Botie tries an inspect-every-machine baseline on four machines. Inspection costs 4 units per machine and a missed faulty machine would cost 30. Calculate inspection costs plus missed-fault costs from the supplied truth labels and decisions.
Confirmed faults, 1 = faulty
  1. 00
  2. 11
  3. 20
  4. 31
Inspection decisions
  1. 01
  2. 11
  3. 21
  4. 31
Cost per inspection
4
Cost per missed fault
30

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.

Which omitted cost or operating constraint could reverse the preference between the rule and model?

Choose a learned model for measured decision benefit under stated assumptions, and preserve the simpler alternative for comparison.

Fault labels are supplied for all pumps. This omits uncertain inspection effectiveness, upkeep, and limited capacity.

Work through the Python challenges →
Continue the courseWhat Can the System Know at Decision Time? →
← What Would This Prediction Actually Change?