Sequential Bayesian decision tool

This tool is designed to help with Sequential Bayesian decision making. You may enter the number of trials for 5 or less variants and number of successes after a test. The tool will calculate the probabilities that each of the variants is the most successful. After each calculation the tool will suggest the split for the next iteration. It can be used as a simulator where the tool will 'home in' on a set of 'hidden probabilities' to find the best.

Example of usage:

  1. Click 'calculate' - (it takes about 5 seconds) - observe graph which appears at base of page.

Example of usage as a simulation:

  1. Click 'simulate' - which fills the successes row with simulated successes using the "hidden" probabilities of success in blue row
  2. Click 'calculate' - observe graph which appears at base of page. Observe probabilities of which is best.
  3. Click 'suggest split' which suggests the numbers to send to the variants in next iteration.
  4. Click 'use suggestion' which fills the numbers for the next iteration at top of tool.
  5. Go to step 1 of this list - continue until exhausted.

Extended example of use for a real set of tests:

  1. Edit names of "variant 1", "variant 2" etc..
  2. Overwrite/edit numbers of trials for each variant.
  3. Overwrite/edit numbers of successes after the trial.
  4. Click 'calculate' - observe graph which appears at base of page.
  5. Observe the probabilities.
  6. Decide how many trials in next iteration, overwrite/edit/leave number in appropriate box.
  7. Click 'suggest split' which suggests the numbers to send to the variants in next iteration.
  8. Click 'use suggestion' which fills the numbers for the next iteration at top of tool.
  9. Go to step 3 of this list

Notes: It can take a long time (10's secs - minutes) with larger numbers. You can zoom (double click, roller wheel) and pan (grab) the graph.
Link to explanatory simulation: here


Editable variant names
Trials this iteration
Hidden probabilities
Successes

Number of iterations so far: Total trials so far:
Total cumulative trials and successes for each variant:


Probabilities that each variant is the most successful:


Number of trials proposed for next iteration:


Cumulative results