OSSForks/tensorflow
1
0
Fork 0
mirror of https://github.com/zebrajr/tensorflow.git synced 2026-01-15 12:15:41 +00:00
Code Issues Actions 5 Packages Projects Releases Wiki Activity
2,461 Commits 50 Branches 144 Tags
8cc567bf9703e14f0e16595eb9f220859a5eff20
Commit Graph

2461 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
Vijay Vasudevan
8cc567bf97 Merge changes from github. 
Change: 123342870
 2016-05-26 12:18:27 -07:00
A. Unique TensorFlower
9a69f398e9 Remove non-legacy public fns in prep for new ones to be added. 
Change: 123341019
 2016-05-26 12:03:50 -07:00
A. Unique TensorFlower
44569014ff Update generated Python Op docs. 
Change: 123339797
 2016-05-26 11:49:08 -07:00
A. Unique TensorFlower
713a82cf8e Adding assign_from_values and assign_from_checkpoint to variables. 
Change: 123339548
 2016-05-26 11:47:23 -07:00
A. Unique TensorFlower
c7462de78d Creates the ReaderSource class, which wraps tensorflow.ReaderBase classes. 
Change: 123336814
 2016-05-26 11:19:01 -07:00
A. Unique TensorFlower
23dc518109 Update generated Python Op docs. 
Change: 123334708
 2016-05-26 11:05:30 -07:00
Benoit Steiner
a1be8581fb Improved the scalability of the cast operation 
Before
BM_cpu_float_int64/64k         90197     201120       7156 8.1GB/s  692.9M items/s
BM_cpu_float_int64/32M      24035828  181447285        100 15.6GB/s  1.3G items/s
BM_cpu_float_half/64k         225119     356610       3046 1.6GB/s  277.6M items/s
BM_cpu_float_half/32M       15364192  144673227        100 12.2GB/s  2.0G items/s
BM_cpu_half_float/64k         216675     352955       3019 1.7GB/s  288.5M items/s
BM_cpu_half_float/32M       21864926  136360857        100 8.6GB/s  1.4G items/s

After
BM_cpu_float_int64/64k         70930     401170      10000 10.3GB/s  881.2M items/s
BM_cpu_float_int64/32M      22222153  225847178        100 16.9GB/s  1.4G items/s
BM_cpu_float_half/64k         163805     259025       4594 2.2GB/s  381.6M items/s
BM_cpu_float_half/32M       15160917  143803486        100 12.4GB/s  2.1G items/s
BM_cpu_half_float/64k         163269     270568       4462 2.2GB/s  382.8M items/s
BM_cpu_half_float/32M       21617823  133221276        100 8.7GB/s  1.4G items/s
Change: 123334623
 2016-05-26 11:03:37 -07:00
Josh Levenberg
1c28846ac8 Follow up to documentation sharding change to make the sharding 
use a fixed hashing function.
Change: 123334435
 2016-05-26 11:03:06 -07:00
A. Unique TensorFlower
ef90ea5cca Adding documentation on integration testing. 
Change: 123334131
 2016-05-26 11:02:36 -07:00
Zongheng Yang
008d41f4cc Implement gradient for SparseSoftmax. 
Essentially the same calculation as Softmax, adapted to sparse inputs/outputs.
Change: 123332988
 2016-05-26 10:49:15 -07:00
A. Unique TensorFlower
6b8de8ac69 In graph_actions, makes sure monitors run only in chief. 
Change: 123332078
 2016-05-26 10:33:20 -07:00
Vijay Vasudevan
e8948a2d9d TensorFlow: Add additional debugging info to error messages when 
a node cannot be placed on a device, specifically when the cause
is due to colocation constraints.

For example, node colocation can cause a group of nodes to be
colocated with each other in an unsatisfiable way.  For example, if we
have three ops, A, B, C, where A supports GPU and CPU, B supports only
GPU, and C supports only CPU, a colocation group has no satisfiable
assignment.  In these cases, the cause is not just the lack of a
kernel for the op that failed to place, but possibly due the set of
ops in the colocation group.

This change adds additional logging to the error message that lists
the op types and their supported devices, so a user can figure out
which combinations of ops are problematic.

Fixes #2508 in that the error message should be clearer now.
Change: 123330848
 2016-05-26 10:32:24 -07:00
A. Unique TensorFlower
afa0112782 Update generated Python Op docs. 
Change: 123329865
 2016-05-26 10:18:20 -07:00
Benoit Steiner
a5b8693369 Made the pooling code fp16 friendly 
Change: 123329817
 2016-05-26 10:17:20 -07:00
A. Unique TensorFlower
a2009115df Update ops-related pbtxt files. 
Change: 123329408
 2016-05-26 10:17:04 -07:00
A. Unique TensorFlower
cafe948be4 Add gradients for batch_cholesky. 
Change: 123328520
 2016-05-26 10:02:25 -07:00
Derek Murray
c1b5075d1f Fix broken Python 3 test: //tensorflow/contrib/learn:graph_io_test 
Change: 123326664
 2016-05-26 09:46:51 -07:00
A. Unique TensorFlower
63013956e3 Update generated Python Op docs. 
Change: 123325658
 2016-05-26 09:33:01 -07:00
Illia Polosukhin
037bfaf05a tf.learn: Adding update op to tf.contrib.layers.optimize_loss to support batch_norm like layers. 
Change: 123324232
 2016-05-26 09:25:40 -07:00
A. Unique TensorFlower
a9112b4a0a Allow the output of the separable_conv2d to have the same number of channels as the corresponding input. 
Change: 123323605
 2016-05-26 09:02:22 -07:00
A. Unique TensorFlower
334618293b Update generated Python Op docs. 
Change: 123322073
 2016-05-26 08:47:58 -07:00
Derek Murray
a6901e9884 Improves shape functions for tf.space_to_batch() and related ops. 
Also fixes tf.batch_to_space(), tf.depth_to_space(), and tf.space_to_depth().

The shape functions now tolerate computed inputs for their paddings argument (whereas before they used the evaluated constant_value, which made them only work with static inputs).

Fixes #2473.
Change: 123320634
 2016-05-26 08:32:54 -07:00
A. Unique TensorFlower
0bb913604a Update generated Python Op docs. 
Change: 123318524
 2016-05-26 08:02:00 -07:00
A. Unique TensorFlower
67782fae2b Update generated Python Op docs. 
Change: 123315764
 2016-05-26 07:23:46 -07:00
A. Unique TensorFlower
c38fbcc6a6 Update generated Python Op docs. 
Change: 123313010
 2016-05-26 06:31:56 -07:00
A. Unique TensorFlower
2767aa0aef Update generated Python Op docs. 
Change: 123310315
 2016-05-26 05:47:16 -07:00
A. Unique TensorFlower
910963836a Update generated Python Op docs. 
Change: 123307673
 2016-05-26 05:02:58 -07:00
A. Unique TensorFlower
9819e968a1 Update generated Python Op docs. 
Change: 123305713
 2016-05-26 04:17:09 -07:00
A. Unique TensorFlower
7eb068ca43 Update generated Python Op docs. 
Change: 123303179
 2016-05-26 03:34:07 -07:00
A. Unique TensorFlower
017175783e Update generated Python Op docs. 
Change: 123300300
 2016-05-26 02:47:07 -07:00
A. Unique TensorFlower
c963048b53 Minor update to example script to produce the reported output. 
Change: 123299440
 2016-05-26 02:32:41 -07:00
A. Unique TensorFlower
9be296aa4b Update generated Python Op docs. 
Change: 123297766
 2016-05-26 02:02:22 -07:00
A. Unique TensorFlower
37adf1eaae Update generated Python Op docs. 
Change: 123295275
 2016-05-26 01:17:13 -07:00
A. Unique TensorFlower
4bf1da6183 Update generated Python Op docs. 
Change: 123293094
 2016-05-26 00:32:04 -07:00
A. Unique TensorFlower
05488ebac7 Update generated Python Op docs. 
Change: 123290850
 2016-05-25 23:48:53 -07:00
A. Unique TensorFlower
3a707b1302 Update generated Python Op docs. 
Change: 123288849
 2016-05-25 23:02:30 -07:00
A. Unique TensorFlower
d29bdd874f Update generated Python Op docs. 
Change: 123286647
 2016-05-25 22:17:13 -07:00
A. Unique TensorFlower
0ef3558007 Update generated Python Op docs. 
Change: 123284722
 2016-05-25 21:32:13 -07:00
A. Unique TensorFlower
d89e30e61f Update generated Python Op docs. 
Change: 123282322
 2016-05-25 20:43:11 -07:00
Xiaoqiang Zheng
8452d29fa8 In case of memory allocation failure, return an error status to stream 
executor.
Change: 123281933
 2016-05-25 20:42:02 -07:00
A. Unique TensorFlower
c41a19137a Update generated Python Op docs. 
Change: 123279971
 2016-05-25 20:02:25 -07:00
Yuan Yu
779c8534b3 Make sure we always add nodes to the session graph for persistent tensors. 
Change: 123279086
 2016-05-25 19:32:29 -07:00
A. Unique TensorFlower
0f79008185 Adding add_loss, get_losses, get_total_loss to loss_ops.py 
Change: 123279011
 2016-05-25 19:31:17 -07:00
Illia Polosukhin
710edb74e6 tf.learn: Added functionality to run over data only once (num_epochs=1). 
Speed up read_batch_example by adding more reading ops before queueing (if reading is slow operation, before it would create a bottleneck on one op).

Speed up read_batch_features by adding parsing_num_threads of parsing ops and queue for aggregating them (currently only one parsing op was used syncronously with training).
Change: 123278248
 2016-05-25 19:17:45 -07:00
A. Unique TensorFlower
2b744876db Update generated Python Op docs. 
Change: 123278130
 2016-05-25 19:16:44 -07:00
A. Unique TensorFlower
041241882c Update generated Python Op docs. 
Change: 123275589
 2016-05-25 18:31:53 -07:00
A. Unique TensorFlower
9c7acc69c1 Update generated Python Op docs. 
Change: 123272356
 2016-05-25 17:49:16 -07:00
A. Unique TensorFlower
23bacdbee6 Fixes bug in SquaredLossUpdater::ComputeUpdatedDual when example weights are different than 1.0. 
Change: 123270998
 2016-05-25 17:31:51 -07:00
Nikhil Thorat
87bd228996 Add run metadata to the mnist_with_summaries.py TF example tutorial. 
This change also changes some of the name_scope's so that the input metadnode is entirely faded for a stats view. Specifically:
- Move the dropout keep_prob scalar into the dropout metanode.
- Move the input reshape into its own metanode. This is run for every step.
Change: 123269596
 2016-05-25 17:16:54 -07:00
A. Unique TensorFlower
858f21884e Update generated Python Op docs. 
Change: 123268244
 2016-05-25 17:02:33 -07:00