Code

Implementing basic RNN#162

Code

Draft

castelao wants to merge 31 commits intomodern-fortran:mainmodern-fortran/neural-fortran:mainfrom

castelao:RNNcastelao/neural-fortran:RNNCopy head branch name to clipboard

Commits

Commits on Sep 15, 2024

• Prototyping RNN layer based on Dense

  Show description for 64e1b69
  castelao

  committed
  64e1b69

  Copy full SHA for 64e1b69

• Extenting uses

  castelao

  committed
  8c11911

  Copy full SHA for 8c11911

• Reading coefficients from h5f model

  Show description for adef7d7
  castelao

  committed
  adef7d7

  Copy full SHA for adef7d7

• feat: get_params()

  castelao

  committed
  b51d66f

  Copy full SHA for b51d66f

• feat: set_params()

  castelao

  committed
  a797502

  Copy full SHA for a797502

• feat: get_num_params()

  castelao

  committed
  ff1c392

  Copy full SHA for ff1c392

• Initializing recurrent kernel and states

  castelao

  committed
  f686950

  Copy full SHA for f686950

• feat: forward()

  castelao

  committed
  acf1afd

  Copy full SHA for acf1afd

• More informative error messages

  castelao

  committed
  69fed32

  Copy full SHA for 69fed32

• Minor adjustments on rnn_layer

  castelao

  committed
  fd24e16

  Copy full SHA for fd24e16

• Constructor for RNN

  castelao

  committed
  7415081

  Copy full SHA for 7415081

• Loading rnn constructor in the root

  castelao

  committed
  6f56863

  Copy full SHA for 6f56863

• Back to 1D concept

  Show description for ad598a8
  castelao

  committed
  ad598a8

  Copy full SHA for ad598a8

• fix: Recurrent is actually a square matrix

  Show description for 0ae7af1
  castelao

  committed
  0ae7af1

  Copy full SHA for 0ae7af1

• Apply loss function if RNN is the output layer

  castelao

  committed
  c164924

  Copy full SHA for c164924

• fix: Getting biases

  castelao

  committed
  55ad96d

  Copy full SHA for 55ad96d

• Allowing backward 1D from dense to RNN

  castelao

  committed
  b345865

  Copy full SHA for b345865

• Allowing backward 1D from RNN

  castelao

  committed
  91b85e0

  Copy full SHA for 91b85e0

• Allowing forward from dense to RNN

  castelao

  committed
  5e197f0

  Copy full SHA for 5e197f0

• Allowing forward from RNN

  castelao

  committed
  7f671c8

  Copy full SHA for 7f671c8

• Getting output from RNN

  castelao

  committed
  c27f59c

  Copy full SHA for c27f59c

• feat: Implementing reset state for RNN

  castelao

  committed
  524d2c4

  Copy full SHA for 524d2c4

• refactor: set_state() on layer level

  Show description for 598f9e7
  castelao

  committed
  598f9e7

  Copy full SHA for 598f9e7

• wip: A simple RNN example

  castelao

  committed
  b7bead6

  Copy full SHA for b7bead6

• feat: layer getting gradient from RNN

  castelao

  committed
  088e4f3

  Copy full SHA for 088e4f3

• feat: layer setting params for RNN

  castelao

  committed
  4d0a4fd

  Copy full SHA for 4d0a4fd

• Might not use set_state at rnn_layer level

  castelao

  committed
  ee516a8

  Copy full SHA for ee516a8

• fix: New access point to 'loss % derivative'

  Show description for 07f7587
  castelao

  committed
  07f7587

  Copy full SHA for 07f7587

• Define set_state as pure

  Show description for 9b22826
  

  castelao

  and

  jvdp1

  committed
  9b22826

  Copy full SHA for 9b22826

• fix: pure interface for set_state

  castelao

  committed
  5bc9bc5

  Copy full SHA for 5bc9bc5

Commits on Oct 21, 2024

• fix: Conciliating with latest main state

  castelao

  committed
  4c7c0b9

  Copy full SHA for 4c7c0b9