Code

Dropout layer#194

Code

Open

milancurcic wants to merge 14 commits intomodern-fortran:mainmodern-fortran/neural-fortran:mainfrom milancurcic:dropoutmilancurcic/neural-fortran:dropoutCopy head branch name to clipboard

+427-16

Commits

Commits on Jan 22, 2025

• First stab at dropout; conflict with base type TODO

  milancurcic

  committed
  e40883b

  Copy full SHA for e40883b

Commits on Jan 23, 2025

• Partial dropout integration

  milancurcic

  committed
  37aa7a5

  Copy full SHA for 37aa7a5

• Test uninitialized dropout layer

  milancurcic

  committed
  820b081

  Copy full SHA for 820b081

• Test dropout state that follows an input layer

  milancurcic

  committed
  75ef184

  Copy full SHA for 75ef184

• Enable forward pass for dropout; backward pass TODO

  milancurcic

  committed
  796ae74

  Copy full SHA for 796ae74

• Version bump and add dropout to the features table

  milancurcic

  committed
  b04d447

  Copy full SHA for b04d447

• Add dropout to CMake

  milancurcic

  committed
  544b23a

  Copy full SHA for 544b23a

Commits on Jan 24, 2025

• Enable preprocessing in fpm.toml (needed with recent versions of fpm)

  milancurcic

  committed
  56dbd52

  Copy full SHA for 56dbd52

• Small change in scale implementation

  milancurcic

  committed
  3b5cc27

  Copy full SHA for 3b5cc27

• Integration of backward pass for dropout

  milancurcic

  committed
  703f802

  Copy full SHA for 703f802

Commits on Feb 6, 2025

• Reduce tolerance in conv2d convergence tests

  milancurcic

  committed
  1dfe6b3

  Copy full SHA for 1dfe6b3

• Fix bug in dropout scaling

  Show description for 59cc7e1
  milancurcic

  committed
  59cc7e1

  Copy full SHA for 59cc7e1

• disable dropout in inference mode (net % predict); TODO enable in net % train

  milancurcic

  committed
  c984b15

  Copy full SHA for c984b15

• Set dropout's training mode to true in net % train(); add tests

  milancurcic

  committed
  e9772a0

  Copy full SHA for e9772a0