Commits

Breadcrumbs

History for

onse-minigo

User selector

All users

All time

Commit History

Commits on Aug 16, 2018

• Script for quantizing weights.

  Show description for f85a685
  

  Ttl

  authored and

  gcp

  committed
  f85a685

  Copy full SHA for f85a685

Commits on Jul 12, 2018

• Fix TF update operations.

  Show description for d0fd3e9
  

  godmoves

  authored and

  gcp

  committed
  d0fd3e9

  Copy full SHA for d0fd3e9

Commits on May 17, 2018

• Multi GPU: fix split and variable placement.

  Show description for 8751123
  

  godmoves

  authored and

  gcp

  committed
  8751123

  Copy full SHA for 8751123

Commits on May 14, 2018

• Add multi GPU training support.

  Show description for 648b230
  

  godmoves

  authored and

  gcp

  committed
  648b230

  Copy full SHA for 648b230

Commits on May 7, 2018

• Make chunkparser robust against malformed data.

  Show description for cd0d852
  gcp

  committed
  cd0d852

  Copy full SHA for cd0d852

Commits on Apr 11, 2018

• Fix dropped bias when widening.

  Show description for 5b2460c
  

  Ttl

  authored and

  gcp

  committed
  5b2460c

  Copy full SHA for 5b2460c

Commits on Apr 3, 2018

• Clean up Python code.

  Show description for 2d2a6a2
  gcp

  committed
  2d2a6a2

  Copy full SHA for 2d2a6a2

• Daemonize TensorFlow child processes.

  Show description for 00a4357
  gcp

  committed
  00a4357

  Copy full SHA for 00a4357

• Support batches larger than GPU RAM size.

  Show description for 7f8ff14
  

  ywrt

  authored and

  gcp

  committed
  7f8ff14

  Copy full SHA for 7f8ff14

Commits on Mar 30, 2018

• net2net: option to increase the number of input planes.

  Show description for 872d4bb
  

  Ttl

  authored and

  gcp

  committed
  872d4bb

  Copy full SHA for 872d4bb

Commits on Mar 26, 2018

• Batch norm beta support.

  Show description for 5a2aca1
  

  Ttl

  authored and

  gcp

  committed
  5a2aca1

  Copy full SHA for 5a2aca1

• Clean up the SWA code.

  Show description for cfed4e4
  

  ywrt

  authored and

  gcp

  committed
  cfed4e4

  Copy full SHA for cfed4e4

Commits on Mar 23, 2018

• Implement stochastic weight averaging.

  Show description for 94f48bb
  

  Ttl

  authored and

  gcp

  committed
  94f48bb

  Copy full SHA for 94f48bb

Commits on Mar 20, 2018

• Fix v2 pipeline: shuffle validation set.

  Show description for 8390055
  gcp

  committed
  8390055

  Copy full SHA for 8390055

Commits on Mar 14, 2018

• Add a unittest for the FileDataSrc class.

  Show description for 66eb43b
  

  ywrt

  authored and

  gcp

  committed
  66eb43b

  Copy full SHA for 66eb43b

• Bugfix FileDataSrc.

  Show description for 7a76eb5
  

  Error323

  authored and

  gcp

  committed
  7a76eb5

  Copy full SHA for 7a76eb5

Commits on Mar 7, 2018

• Fix training to work on Windows again.

  Show description for ee5f391
  

  ywrt

  authored and

  gcp

  committed
  ee5f391

  Copy full SHA for ee5f391

Commits on Mar 2, 2018

• Move ChunkParser into a seperate file.

  Show description for 3bf9a7f
  

  ywrt

  authored and

  gcp

  committed
  3bf9a7f

  Copy full SHA for 3bf9a7f

Commits on Feb 28, 2018

• Update TensorFlow pipeline.

  Show description for 800990c
  

  ywrt

  authored and

  gcp

  committed
  800990c

  Copy full SHA for 800990c

Commits on Feb 6, 2018

• TensorBoard: capitalize the L in "Policy Loss".

  Show description for 0e8f1c6
  

  FFLaguna

  authored and

  gcp

  committed
  0e8f1c6

  Copy full SHA for 0e8f1c6

Commits on Jan 31, 2018

• Do not regularize fully-connected layer biases.

  Show description for 99119bb
  gcp

  committed
  99119bb

  Copy full SHA for 99119bb

Commits on Jan 30, 2018

• Increase skip factor for chunk parsing.

  Show description for ef66e26
  gcp

  committed
  ef66e26

  Copy full SHA for ef66e26

• Speed up chunk parsing/data split.

  Show description for 9e9fd49
  gcp

  committed
  9e9fd49

  Copy full SHA for 9e9fd49

• Add the MongoDB training data dumper.

  Show description for 1564e85
  gcp

  committed
  1564e85

  Copy full SHA for 1564e85

• Make net_to_model work again, print policy loss.

  Show description for 1559ada
  gcp

  committed
  1559ada

  Copy full SHA for 1559ada

• Use a feedable iterator rather than tf.cond().

  Show description for 012bd8c
  gcp

  committed
  012bd8c

  Copy full SHA for 012bd8c

• Let chunk parsing skip every 1/8 positions.

  Show description for b43f29f
  gcp

  committed
  b43f29f

  Copy full SHA for b43f29f

• Dump supervised learning per game. Shuffle chunks.

  Show description for d450e8e
  gcp

  committed
  d450e8e

  Copy full SHA for d450e8e

• Implement real validation split for TF.

  Show description for d396c6a
  gcp

  committed
  d396c6a

  Copy full SHA for d396c6a

Commits on Jan 24, 2018

• Update requirements.txt.

  Show description for cb7e22f
  gcp

  committed
  cb7e22f

  Copy full SHA for cb7e22f

• Add net2net.py for enlarging networks.

  Show description for 6f93c0e
  

  Ttl

  authored and

  gcp

  committed
  6f93c0e

  Copy full SHA for 6f93c0e

Commits on Jan 23, 2018

• Check that filters/blocks match in net_to_model.

  Show description for a3a2e0f
  

  Ttl

  authored and

  gcp

  committed
  a3a2e0f

  Copy full SHA for a3a2e0f

• Update default TensorFlow parameters.

  Show description for cabe7d1
  gcp

  committed
  cabe7d1

  Copy full SHA for cabe7d1

Commits on Jan 9, 2018

• Scale MSE loss back up in reported total loss.

  Show description for b6a87da
  gcp

  committed
  b6a87da

  Copy full SHA for b6a87da

Commits on Jan 4, 2018

• Reduce training verbosity.

  Show description for 5e9bcd4
  gcp

  committed
  5e9bcd4

  Copy full SHA for 5e9bcd4