Prior Beliefs and New Evidence
The service has a historical late-delivery rate before today's forecast arrives.
Prior Beliefs and New Evidence explained visually
The service has a historical late-delivery rate before today's forecast arrives. English captions are optional and start off. The complete transcript follows the written lesson.
What you will learn
The service has a historical late-delivery rate before today's forecast arrives.
A practical question
Before seeing today's forecast, the dispatcher already knows that some fraction of trips run late. That baseline is a prior belief about a delivery under a defined set of conditions.
The useful idea
A weather alert changes the picture only in relation to how often alerts occur on late trips and on-time trips. If alerts are common on both, one alert may not move the risk very far. The same alert can lead to a different final risk when the baseline late rate differs by route.
Work through the example
Compare an alert for a historically reliable route with an alert for a difficult one, using held-out frequency estimates. Make sure the prior and alert statistics come from comparable trips. A prior is evidence summarized from a population, not a personal feeling about one driver.
Limits and next step
Next, we'll work through the counts in a concrete Bayes example so the size of the update is clear.
Worked example
Update a low base rate using a weather signal rather than treating that signal alone as the answer.
Quick check: test your understanding
Try it yourself: In the situation described above, explain the method and one reason its output might be misleading.
Check the reasoning
A prior expresses an initial probability before the new evidence is considered. Ignoring the base rate can exaggerate a warning.
What is the main idea behind prior beliefs and new evidence?
A prior expresses an initial probability before the new evidence is considered.
How is it used in the example?
Update a low base rate using a weather signal rather than treating that signal alone as the answer.
What limitation should be checked?
Ignoring the base rate can exaggerate a warning.
Further reading
Machine Learning — Tom M. Mitchell.
Original ToolsInVision teaching examples. Editorial approach · Suggest a correction
Accessible referenceLesson 29 transcript147 words · Video transcript · Matches the published narration and captionsOpen transcript
Before seeing today's forecast, the dispatcher already knows that some fraction of trips run late. That baseline is a prior belief about a delivery under a defined set of conditions.
A weather alert changes the picture only in relation to how often alerts occur on late trips and on-time trips. If alerts are common on both, one alert may not move the risk very far. The same alert can lead to a different final risk when the baseline late rate differs by route.
Compare an alert for a historically reliable route with an alert for a difficult one, using held-out frequency estimates. Make sure the prior and alert statistics come from comparable trips. A prior is evidence summarized from a population, not a personal feeling about one driver.
Next, we'll work through the counts in a concrete Bayes example so the size of the update is clear.