Code

Visualization#1

Code

Open

roys174 wants to merge 63 commits intoNoahs-ARK:masterNoahs-ARK/rational-recurrences:masterfrom dodgejesse:visualizationdodgejesse/rational-recurrences:visualizationCopy head branch name to clipboard

+3,475-117

Commits

Commits on Oct 12, 2018

• first attempt at group lasso

  

  Jesse Dodge

  committed
  f906bc5

• experimenting with stuff

  dodgejesse

  committed
  5a42380

Commits on Oct 17, 2018

• added random hyperparameter search, and logging.

  dodgejesse

  committed
  22b739e

• removed old running script

  dodgejesse

  committed
  66ed540

• manually edited files to match explore brach of private repo

  dodgejesse

  committed
  ecc878f

Commits on Oct 19, 2018

• added n-gram cpu computation

  dodgejesse

  committed
  9d6d3e6

Commits on Oct 23, 2018

• added cuda implementation of fourgram forward and backward computation

  dodgejesse

  committed
  ae50123

Commits on Oct 26, 2018

• fourgram model works, can use two types of regularization.

  dodgejesse

  committed
  53d1336

• implemented 1,2,3,4 gram models in cuda.

  dodgejesse

  committed
  14aee6e

Commits on Oct 27, 2018

• can now train models which can capture 1,2,3,4 grams, aka learned structures

  dodgejesse

  committed
  e9b8b9e

Commits on Nov 8, 2018

• small changes

  dodgejesse

  committed
  5e897bf

Commits on Nov 24, 2018

• runs on amazon_categories. implemented entropy regularization. can load models learned with entropy regularization.

  dodgejesse

  committed
  8583d59

Commits on Dec 4, 2018

• adding more hyperparameter optimization

  dodgejesse

  committed
  ac08ea4

Commits on Dec 6, 2018

• currently stuck on training small learning rate, gradient clipping, rho_entropy, as it wont get above .9 after 50 epochs.

  dodgejesse

  committed
  a948e8e

Commits on Dec 8, 2018

• now searches over regularization strength, and resamples if the learning rate is too high or low.

  dodgejesse

  committed
  dea3c88

• small changes

  dodgejesse

  committed
  6e19681

Commits on Dec 13, 2018

• small changes

  dodgejesse

  committed
  fa8b8eb

Commits on Dec 19, 2018

• added ability to load learend group norms (for l1 regularization), and pulled out the entropy regularization experiment code to its own module

  dodgejesse

  committed
  da4a077

Commits on Dec 21, 2018

• small changes

  dodgejesse

  committed
  9fd0705

• consolidated regularization search into one script.

  dodgejesse

  committed
  22e9719

Commits on Dec 22, 2018

• added proximal gradient descent, though it doesnt work yet. its currently updating a copy of the weights instead of the weights themselves.

  dodgejesse

  committed
  297e7a2

Commits on Dec 27, 2018

• hacked together proximal gradient descent. seems to work.

  dodgejesse

  committed
  8bca89c

Commits on Dec 28, 2018

• added another check for regularization search. just using regularization added to loss now, not proximal gradient

  dodgejesse

  committed
  12f0da6

Commits on Jan 3, 2019

• small changes

  dodgejesse

  committed
  398b3f8

Commits on Jan 16, 2019

• small changes

  dodgejesse

  committed
  052bb60

Commits on Jan 17, 2019

• small changes

  dodgejesse

  committed
  48b65e6

• combined the code to load learned structures into one module

  dodgejesse

  committed
  7b446d2

• small changes

  dodgejesse

  committed
  de4d2bd

• small changes

  dodgejesse

  committed
  c9dd423

Commits on Jan 20, 2019

• 1. Added local version of experiment (w/o hyperparameter search). 2. Added parameters to experiment_params.py. 3. Added option to save model

  

  Roy Schwartz

  committed
  9a424a5

Commits on Jan 21, 2019

• Added option to read trained model

  

  Roy Schwartz

  committed
  1c69029

• Refactored code to work with input arguments

  

  Roy Schwartz

  committed
  da3b5d7

• Fixed writing model command, removed running on test during training

  

  Roy Schwartz

  committed
  346785f

• General tools for RR

  

  Roy Schwartz

  committed
  4e9a42a

• Fixed writing model command, removed running on test during training

  

  Roy Schwartz

  committed
  d381441

• Made a couple of arguments required

  

  Roy Schwartz

  committed
  313fb08

• Reverted wrong commit

  

  Roy Schwartz

  committed
  d78065e

Commits on Jan 23, 2019

• Running embedding dropout only at train time

  

  Roy Schwartz

  committed
  ca8cbba

• added simple example experiment with known results to run_current_experiment.py

  dodgejesse

  committed
  1f4062c

• Embedding dropout fix is redundant

  

  Roy Schwartz

  committed
  014c9ce

Commits on Jan 28, 2019

• Default seed value (for reproducible experiments)

  

  Roy Schwartz

  committed
  1461373

• Visualization code: keep trace of max score when computing forward in rrnn.py, then print top scoring experssions for each pattern (only works for CPUs)

  

  Roy Schwartz

  committed
  fd8b4e3

• Merge branch 'visualization' of https://github.com/dodgejesse/rational-recurrences into visualization

  

  Roy Schwartz

  committed
  bdb9e39

• Remove debug prints

  

  Roy Schwartz

  committed
  38d2a78

• Visualization bug fixes

  

  Roy Schwartz

  committed
  224ae19

• Merge branch 'visualization' of https://github.com/dodgejesse/rational-recurrences into visualization

  

  Roy Schwartz

  committed
  f11dad3

• Fixed visualization bug that occurs if a specific pattern length has no patterns in practice

  

  Roy Schwartz

  committed
  59086b2

• Cleaned code up a bit, fix score calculation code

  

  Roy Schwartz

  committed
  3b9def7

• Merge branch 'visualization' of https://github.com/dodgejesse/rational-recurrences into visualization

  

  Roy Schwartz

  committed
  2f65408

• forget gate set to 0 when i=0

  

  Roy Schwartz

  committed
  880aed8

• Merge branch 'visualization' of https://github.com/dodgejesse/rational-recurrences into visualization

  

  Roy Schwartz

  committed
  4838ef0

• Correctly printing pattern length in summary

  

  Roy Schwartz

  committed
  a510da2

• Added -inf as score of unreachable states in visualization

  

  Roy Schwartz

  committed
  f7b97ac

• Merge branch 'visualization' of https://github.com/dodgejesse/rational-recurrences into visualization

  

  Roy Schwartz

  committed
  ff73f20

Commits on Jan 29, 2019

• Removed a few old commented code

  

  Roy Schwartz

  committed
  6a4c68c

Commits on Jan 31, 2019

• 1. Fixed bug that made code only run on last batch. 2. Added best matches for intermediate states in pattern. 3. Added some documentation. 4. Initial implementation of parallel patterns (not tested…

  Show description for 179b72f
  

  Roy Schwartz

  committed
  179b72f

• Parralized visualization, now computes it for all patterns in parallel

  

  Roy Schwartz

  committed
  8638c82

Commits on Feb 1, 2019

• Support for use_last_cs argument in command line

  

  Roy Schwartz

  committed
  5ac0631

• Visualization now works in parallel. Also added some documentation

  

  Roy Schwartz

  committed
  92cbbe1

• Merge branch 'visualization' of https://github.com/dodgejesse/rational-recurrences into visualization

  

  Roy Schwartz

  committed
  7e0d4e8

Commits on Feb 2, 2019

• Printing matches with lowest scores as well

  

  Roy Schwartz

  committed
  811958a

Commits on Feb 6, 2019

• Only visualize pattern states that are of high norm

  

  Roy Schwartz

  committed
  50a25db

• Merge branch 'visualization' of https://github.com/dodgejesse/rational-recurrences into visualization

  

  Roy Schwartz

  committed
  4832d7b