Code

Tuner improvements and fixes#55

Code

Merged

Dannyj1 merged 9 commits intomasterDannyj1/Zagreus:masterfrom tuner-improvementsDannyj1/Zagreus:tuner-improvementsCopy head branch name to clipboardNov 26, 2023

+127-74

Commits

Commits on Nov 25, 2023

• Establish a baseline to compare improvements against

  Show description for f97bdd9
  Dannyj1

  committed
  f97bdd9

  Copy full SHA for f97bdd9

• Fix inaccuracies in the loss function

  Show description for 402f4dd
  Dannyj1

  committed
  402f4dd

  Copy full SHA for 402f4dd

• Merge branch 'master' into tuner-improvements

  Dannyj1

  authored
  d6a2323

  Copy full SHA for d6a2323

• Apply rounding when going from gradients to eval parameters instead of just casting to int

  Show description for f54c78d
  Dannyj1

  committed
  f54c78d

  Copy full SHA for f54c78d

• Revert "Apply rounding when going from gradients to eval parameters instead of just casting to int"

  Show description for 1024ab5
  Dannyj1

  committed
  1024ab5

  Copy full SHA for 1024ab5

• Properly implement batching

  Show description for 7832c41
  Dannyj1

  committed
  7832c41

  Copy full SHA for 7832c41

• Use a moving average for early stopping, change gradient estimation method

  Show description for c9910ed
  Dannyj1

  committed
  c9910ed

  Copy full SHA for c9910ed

• Use a different tuner dataset

  Show description for 95b17ce
  Dannyj1

  committed
  95b17ce

  Copy full SHA for 95b17ce

Commits on Nov 26, 2023

• Revert "Use a different tuner dataset"

  Show description for b104e0b
  Dannyj1

  committed
  b104e0b

  Copy full SHA for b104e0b