Commits

Branch selector

 master

User selector

All users

All time

Commit History

Commits on Aug 6, 2018

• Merge pull request #34 from PPPLDeepLearning/reduce_options

  Show description for 731d975
  jnkh

  authored
  731d975

  Copy full SHA for 731d975

• Update mpi_runner.py

  Show description for ea5b3e6
  jnkh

  authored
  ea5b3e6

  Copy full SHA for ea5b3e6

• Do not broadcast weights after all-reduce

  ASvyatkovskiy

  committed
  2bff1ef

  Copy full SHA for 2bff1ef

Commits on Mar 13, 2018

• Merge pull request #33 from PPPLDeepLearning/import_fixes_py2

  Show description for 92f86f2
  ASvyatkovskiy

  authored
  92f86f2

  Copy full SHA for 92f86f2

Commits on Mar 12, 2018

• Fix t.MaxHingeTarget.fac -> MaxHingeTarget.fac

  ASvyatkovskiy

  committed
  66cf3f8

  Copy full SHA for 66cf3f8

• Defer import causing a cyclic dependency until later. Only needed in Python 2

  ASvyatkovskiy

  committed
  d4270cd

  Copy full SHA for d4270cd

• Defer import causing a cyclic dependency until later. Only needed in Python 2

  ASvyatkovskiy

  committed
  48e314d

  Copy full SHA for 48e314d

• Bug fix: proper plasma.conf import

  ASvyatkovskiy

  authored
  541d98e

  Copy full SHA for 541d98e

• Import error in targets

  ASvyatkovskiy

  authored
  5aa60ed

  Copy full SHA for 5aa60ed

• Merge pull request #32 from PPPLDeepLearning/early_stopping_fixes

  Show description for 2e89ffa
  ASvyatkovskiy

  authored
  2e89ffa

  Copy full SHA for 2e89ffa

Commits on Mar 11, 2018

• Relax Keras version requirement now

  ASvyatkovskiy

  committed
  7484c11

  Copy full SHA for 7484c11

• After 2.0.6, keras Model does not have stop_training attribute. Only added in the callback

  ASvyatkovskiy

  committed
  e42ba53

  Copy full SHA for e42ba53

Commits on Feb 23, 2018

• Merge pull request #30 from PPPLDeepLearning/loss_scaling

  Show description for 15613fa
  ASvyatkovskiy

  authored
  15613fa

  Copy full SHA for 15613fa

• Use scratch gpfs on Tigergpu - SSDs

  ASvyatkovskiy

  committed
  151738c

  Copy full SHA for 151738c

• Add loss scaling option to targets

  ASvyatkovskiy

  committed
  210ede8

  Copy full SHA for 210ede8

• Unscale gradients by DUMMY_LR factor before subtracting, unscale loss…

  ASvyatkovskiy

  committed
  d44a2f6

  Copy full SHA for d44a2f6

• Add a config parameter to control loss scaling. Default is 1.0

  ASvyatkovskiy

  committed
  9aded7f

  Copy full SHA for 9aded7f

Commits on Feb 9, 2018

• Merge pull request #29 from PPPLDeepLearning/jdev

  Show description for 8512cd9
  jnkh

  authored
  8512cd9

  Copy full SHA for 8512cd9

• modified printout to add walltime

  

  Julian Kates-Harbeck

  committed
  e21f431

  Copy full SHA for e21f431

• removed print

  

  Julian Kates-Harbeck

  committed
  e06bcb1

  Copy full SHA for e06bcb1

• implemented randomized cutting of first batch_size shots in the buffer. This means shots end at random times even at the beginning of training, which removes the bias that for the first couple of t…

  Show description for b080734
  

  Julian Kates-Harbeck

  committed
  b080734

  Copy full SHA for b080734

• conf update

  

  Julian Kates-Harbeck

  committed
  e5c3f1b

  Copy full SHA for e5c3f1b

• conf update

  

  Julian Kates-Harbeck

  committed
  dea576e

  Copy full SHA for dea576e

• not saving from within preprocessor class

  

  Julian Kates-Harbeck

  authored and

  Julian Kates-Harbeck

  committed
  374e069

  Copy full SHA for 374e069

• not saving from within preprocessor class

  

  Julian Kates-Harbeck

  authored and

  Julian Kates-Harbeck

  committed
  480e303

  Copy full SHA for 480e303

• merged

  

  Julian Kates-Harbeck

  authored and

  Julian Kates-Harbeck

  committed
  1827332

  Copy full SHA for 1827332

• fixed bug where all workers had the same numpy random seed (it gets reset to 0 in preprocessor). Thus need to explicitly give each worker a different random seed right before training.

  

  Julian Kates-Harbeck

  committed
  8ce98dc

  Copy full SHA for 8ce98dc

Commits on Feb 7, 2018

• backwards compatibility for normalizer (check if signal has is_positive attribute). Plotting options for performance analyzer

  

  Julian Kates-Harbeck

  committed
  85d4b8b

  Copy full SHA for 85d4b8b

• Merge branch 'jdev' of https://github.com/PPPLDeepLearning/plasma-python into jdev

  

  Julian Kates-Harbeck

  committed
  8ecb9a4

  Copy full SHA for 8ecb9a4

• recasting to shotlist

  

  Julian Kates-Harbeck

  committed
  b6c0f3d

  Copy full SHA for b6c0f3d

• added saving of ROC curves and accumulated disruptions

  

  Julian Kates-Harbeck

  authored and

  Julian Kates-Harbeck

  committed
  5052845

  Copy full SHA for 5052845

Commits on Feb 5, 2018

• Merge branch 'jdev' of https://github.com/PPPLDeepLearning/plasma-python

  

  Julian Kates-Harbeck

  authored and

  Julian Kates-Harbeck

  committed
  69063c0

  Copy full SHA for 69063c0

• latest version of scaling plot notebook

  

  Julian Kates-Harbeck

  authored and

  Julian Kates-Harbeck

  committed
  8f4c7d1

  Copy full SHA for 8f4c7d1

Commits on Feb 1, 2018

• higher numbers of conv filters in tune hyperparams

  

  Julian Kates-Harbeck

  committed
  e22b32e

  Copy full SHA for e22b32e

Commits on Jan 30, 2018

• Roll back to Keras 2.0.6. Need to investigate changes to model.stop_training in Keras 2.1.x

  ASvyatkovskiy

  authored
  77e41a5

  Copy full SHA for 77e41a5