Estimate the start time by variance

[kkdlau]

This PR is for fixing #257

Below is a summary of the changes made in this pull request:

• New OnlineCalculator classes - WelfordVarianceCalculator and OnlineStatisticsCalculator
  • This two classes allows to compute the mean and variance on-the-fly, without caching the dataset in memory
• Test cases for OnlineCalculator classes
• Update the way of computing estimated start time

Regarding the estimated start time, I used a slightly different approach than suggested in #257

• GC frequency is calculated as the time gap between two consecutive GC events.
  • For example, if event A happens at 2s and event B happens at 2.3s, the GC frequency is 2.3s - 2s = 0.3s.
• If timeOfFirstEvent does not have time stamp, the method will return date stamp - SD of GC frequency
• If timeOfFirstEvent has time stamp
• it will first compute time stamp - SD of GC frequency
  • if the value is greater than 0, then return value directly
  • otherwise return timeOfFirstEvent, since we cannot have negative time stamp

Please review the code and provide any comments or feedback you may have. I would also appreciate clarification on the specific points mentioned above!

[kkdlau]

@microsoft-github-policy-service agree

[karianna]

Hi @kkdlau - @kcpeppe sees this as a good change :-), are you able to go ahead and also update the tests to match?

[kkdlau]

Hi @kkdlau - @kcpeppe sees this as a good change :-), are you able to go ahead and also update the tests to match?

definitely! Will push the change later

[karianna]

This does seem incorrect? It's a duplicate?