hash stringlengths 40 40 | msg stringlengths 1 131k | author stringlengths 1 33 | email stringlengths 0 57 | date int64 1,447B 1,698B |
|---|---|---|---|---|
cb41260fec9aae129e2378cc2ddd7d128ec2ad23 | Replace javascript function name `isNodeAuxilliary` with `isNodeAuxiliary` | Dongjoon Hyun | dongjoon@apache.org | 1,452,295,579,000 |
e7ca22d689813417fda19b9bc4e12628fc3d5a69 | Add dependency on paper_tabs to TensorBoard
Change: 111713364 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,297,607,000 |
13a4ddafc5e7a0211c23e2f488a686c2408aacd9 | Added clear_stop() flag to the coordinator.
Change: 111714387 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,297,615,000 |
771a92208fa2301ca1d54ec53b8bfc7d227e3d66 | Fix minor typos in example code and docs.
Change: 111719502 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,297,624,000 |
67eb676b1044dd6fa883b45bc163366ecd3ce646 | Fixed doc string for switch()
Change: 111724690 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,297,635,000 |
4e68ec99872f21e4a401d868b2656ca26b56981a | Add support for allocators that depend on the state of
the local resource_manager. This is needed for experimental device
support.
Change: 111729000 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,297,644,000 |
cd7e98bde1107b0ba39695fded228b98563073b7 | Added control dependencies on grad to prevent ops from being computed too early.
Change: 111730245 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,297,652,000 |
7116cc8b39c432f64e6e82b5708d2773f3fec159 | Register tf.add() for strings, with concatenation (and broadcasting) semantics.
This matches the behavior of numpy, for arrays of objects:
>>> x = np.array(["foo", "bar"], dtype=np.object)
>>> y = np.array(["baz", "quux"], dtype=np.object)
array(['foobaz', 'barquux'], dtype=object)
Change: 111732290 | Derek Murray | mrry@google.com | 1,452,297,668,000 |
8e74f73fab39f5e5bea1abc08bbd16c097351e55 | Rollback of "Added control dependencies on grad to prevent ops from being computed too early."
Change: 111736819 | Vijay Vasudevan | vrv@google.com | 1,452,297,677,000 |
c7cf1a972e41bf2acb5f12cf98fd215ba3fcb0ba | Merge commit for internal changes | Vijay Vasudevan | vrv@google.com | 1,452,297,781,000 |
22267addd64b00f9457605cdce3d82276a478780 | Rollback of "Pulled the latest version of Eigen, which offers much faster reductions on GPU"
Change: 111739152 | Vijay Vasudevan | vrv@google.com | 1,452,299,024,000 |
740b5a207335dcc21e95a43e9e87c4d7af728312 | Merge commit for internal changes | Vijay Vasudevan | vrv@google.com | 1,452,302,736,000 |
f9514a917265f0e98c8fb44abc51158685f72ca6 | Merge pull request #731 from dongjoon-hyun/rename_js_function_isNodeAuxilliary
Replace javascript function name `isNodeAuxilliary` with `isNodeAuxiliary` | Daniel Smilkov | dsmilkov@gmail.com | 1,452,312,328,000 |
8e8ad2d409e2fc07116c07276e9923e2266c9eb7 | * Temporarily omit some io stuff from coverage | Yuan (Terry) Tang | terrytangyuan@users.noreply.github.com | 1,452,396,626,000 |
a666ccdd73e63369d90ba37073a9e05345ea42e8 | Fix for logic in re-creating summary writer on subsequent fit() | Illia Polosukhin | ilblackdragon@gmail.com | 1,452,407,117,000 |
67acf27a61bf030f1afc4d7185a54605ae0d1118 | Switching to use tf.nn.rnn_cell.linear instead of importing. This will work in tensorflow >= 0.6 | Illia Polosukhin | ilblackdragon@gmail.com | 1,452,407,437,000 |
21df765ae4a35f655b068ae7d2440da98944cb7e | Adding rnn_decoder with sampling sub-graph. Adding rnn seq2seq that will work with it. | Illia Polosukhin | ilblackdragon@gmail.com | 1,452,407,523,000 |
8ddbadb408e2eb90e0ef8493054e566260e2e38a | Merge branch 'master' of https://github.com/tensorflow/tensorflow | Kenton Lee | kenton.ct.lee@gmail.com | 1,452,468,228,000 |
6a0488f2c102c93a6487798dcf21be6c38916a27 | Fix unmatched html tags | Dongjoon Hyun | dongjoon@apache.org | 1,452,298,106,000 |
b5e1874e7f3fc879772576fa64a4becf86baf53c | Merge pull request #732 from dongjoon-hyun/fix_unmatched_html_tags. Closes #732. | Vijay Vasudevan | vrv@google.com | 1,452,490,831,000 |
400ec92634415ab4e982c42ee183be9104e46d3d | Fix typos in tensorboard/components | Dongjoon Hyun | dongjoon@apache.org | 1,452,295,444,000 |
cfc73af6e2ed4a5ccfd259aeb1781dab4fad7292 | Merge pull request #735 from dongjoon-hyun/fix_dtypos_in_tensorboard_components. Closes #735. | Vijay Vasudevan | vrv@google.com | 1,452,491,588,000 |
22b6b23d9778fe7ef53e1597248558f21d1dda48 | Merge pull request #541 from kentonl/master
Allows IndexedSlices to be fed and fetched. | Derek Murray | derek.murray@gmail.com | 1,452,534,185,000 |
e7f63369e74e6500e27cff451c84623386fd05a8 | Fix broken reference to TensorBase::Serialize | Dan Vanderkam | danvdk@gmail.com | 1,452,539,959,000 |
f566060e36577ea5a57c1b677fd5705c7d3935c1 | Merge pull request #746 from danvk/master. Closes #746 | Vijay Vasudevan | vrv@google.com | 1,452,540,539,000 |
492e55ff2297dc13f5ed540056ad40603f2b1dd0 | Add new parse_example functions that take namedtuple config objects, and prepend old API functions with "legacy_".
Change: 111745511 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,540,994,000 |
60bccf654abc869b19b2ec58d021421ef9480445 | Drop macros.h from port.h now that everything that needs
macros.h #includes it directly.
Change: 111775240 | Josh Levenberg | josh11b@tensorflow.org | 1,452,541,003,000 |
a6037e933fe8799c80f94decc9f5b178b833c1b8 | Add better shape inference for `tf.zeros_like()` and `tf.ones_like()`.
Previously, partial shape information was discarded, because our
constant evaluation for (e.g.) `tf.shape(tf.placeholder([..., None,
...]))` could not produce a Numpy array for the shape. Since the
*_like wrappers have access to the input tensor, we can use
`Tensor.set_shape()` to add back the partial information.
Fixes #744.
Change: 111856452 | Derek Murray | mrry@google.com | 1,452,541,021,000 |
c1e0803e8419282f11136e46dadeb66bae0b9d13 | Small change in wording.
Change: 111861092 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,541,039,000 |
681db2caf1fd48dfc85500817cd40cfde78dc67a | Add note regarding when we can make parsing of AttrValue's holding an
empty list more strict. We had some convenient GraphDef version bumps
at the relevant time.
Change: 111863523 | Josh Levenberg | josh11b@tensorflow.org | 1,452,541,047,000 |
96aa23e45e72bd765920cc957090ff453e25cf64 | TensorFlow: remove simple executor state and has_control_flow flag everywhere.
Change: 111871361 | Vijay Vasudevan | vrv@google.com | 1,452,541,057,000 |
1b94cb11a97186a27cf8be5085fa85d3f45ba355 | fix misspells. | Fabrizio (Misto) Milo | mistobaan@gmail.com | 1,452,537,735,000 |
b45df4534d9d6101a76775206798dd9c6e44dd8d | Merge pull request #745 from Mistobaan:feature/fix-small-misspells. Closes #745 | Vijay Vasudevan | vrv@google.com | 1,452,541,478,000 |
5f0db85b7d1f70eb4128fb7dbaa7eae2118d159b | Fix TensorBoard not displaying any images | Bart Coppens | kde@bartcoppens.be | 1,452,541,862,000 |
38f54b55cdc66e12a2f6ce9d6724440e9738ac6f | Upgraded to the latest version of Eigen, which fixes a compilation error as
well the crash reported in https://github.com/tensorflow/tensorflow/issues/713.
Change: 111874622 | Benoit Steiner | benoit.steiner.goog@gmail.com | 1,452,546,209,000 |
cc30784d0ff5ae81e2cf75a4e081b0e2153456f8 | Change: 111879963 | Vijay Vasudevan | vrv@google.com | 1,452,546,218,000 |
81991e26e1ad5a5eb8cec5d53d9490b25dc4079c | Fix typos in tensorboard and tools | Dongjoon Hyun | dongjoon@apache.org | 1,452,546,487,000 |
b9db3b3d4a98acc9681b7d32f835fcf45686f04b | Change: 111880705 | Vijay Vasudevan | vrv@google.com | 1,452,546,659,000 |
9a27fb169c5b6eabc1c5ce490bef2fa1c7b608e3 | Merge pull request #748 from dongjoon-hyun/fix_typos_in_tensorboard_and_tools. Closes #748 | Vijay Vasudevan | vrv@google.com | 1,452,546,699,000 |
d1b8333effdcb031e6e34a2835a2f1c877fdd79b | Merge commit for internal changes | Vijay Vasudevan | vrv@google.com | 1,452,546,810,000 |
5853ad99dd3f8b443fdb97d83de35fe057db9e7f | TensorFlow: fix importer to properly fill in default values.
Accessing the attr of a node when not present creates a proto3 default
which has an empty value structure. So checking for None was not
sufficient: we need to check whether it has a value defined in it,
and if not, to copy the default value into it.
Change: 111887578 | Vijay Vasudevan | vrv@google.com | 1,452,576,391,000 |
39ac7343cf47add2c78d9fe4792bdcaedb99affc | Adds support for SQRTN combiner.
The implementation divides the weighted sum by sqrt(n) or divides by the sqrt(sum(x^2)) if sp_weights are specified on embedding_lookup_sparse.
Also implements math_ops.sparse_segment_sqrtn and its corresponding math_ops.sparse_segment_sqrtn_grad.
Change: 111889513 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,576,402,000 |
f2de334678c251aa767743d08e66345a45113cc4 | Fix type handling logic in translation of TensorFlow functions.
Inputs and outputs with lists of fixed types (as opposed to polymorphic lists) caused incorrect translation of some nodes in a function graph - adding a spurious attribute with name == "" and no value.
e.g. the 'indices' argument of DynamicStitch has a fixed int32 type, but 'data' is polymorphic. Handling 'indices' added a bogus attribute to the op in the function body.
REGISTER_OP("DynamicStitch")
.Input("indices: N * int32")
.Input("data: N * T")
.Output("merged: T")
.Attr("N : int >= 2")
.Attr("T : type")
Note that this is because the attr map in the ArgDef behaves like a C++ map in that an attempt to access a non-existent field creates key with a default value (i.e. does not fail like a Python dictionary would).
Change: 111895250 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,576,410,000 |
5b598e593128d2d8aedf50070f4bf6362de465d6 | Catch missing session error earlier when queue runners get called without default session
Change: 111896446 | Yaroslav Bulatov | yaroslavvb@google.com | 1,452,576,418,000 |
277184ba512226a11bb3c75093b3eab4bf8aac1d | Add support for storing Partial TensorShapes in python & op-def code.
This includes:
* Modify python's TensorShape to allow serialization of partially defined shapes.
* Update attr related code to support "partial_shape" and "list(partial_shape)"
declarations.
* Link partial_shape attr declarations to C++ PartialTensorShape object.
Change: 111902026 | Eugene Brevdo | ebrevdo@gmail.com | 1,452,576,427,000 |
339186c1a9606bc0d333e61e314c2db12b4ac24f | Reduce the warning size for optional scratch space allocation failure.
Change: 111907573 | Xiaoqiang Zheng | zhengxq@google.com | 1,452,576,435,000 |
c08d2206beb3ac3187a3ea14777c07bda012610b | Merge commit for internal changes | Vijay Vasudevan | vrv@google.com | 1,452,576,508,000 |
3034eef57a9e6a080eae457a5ee8480fadd377ea | Fix error in link to mnist_with_summaries.py
Change: 111920304 | Martin Wicke | wicke@google.com | 1,452,614,789,000 |
a3ebb5f334d8c2c35de423af30b77b750ebec996 | Drop mutex.h from port.h now that everything that needs
mutex.h #includes it directly.
Change: 111921996 | Josh Levenberg | josh11b@tensorflow.org | 1,452,614,798,000 |
ca34dbe4e8f3fe6fc1e0e209662f93afe4ff9614 | Update generated Op docs.
Change: 111922002 | Yutaka Leon | yutaka.leon@gmail.com | 1,452,614,807,000 |
cbff45c03b96a2efb2bd45a54cd0facb12007fa8 | Add GraphDef version 6 implementing scalar strictness, but don't use it
The kAllowLegacyScalars flag is now gone. Instead, scalar strictness now
depends on the GraphDef version: we are lenient below 6 and strict with 6
and above. The current GraphDef version is still 5; new graphs will not
yet use the new version by default.
Outside of PLATFORM_GOOGLE, this change has no effect since the code was
already scalar strict.
The TensorShapeUtils versions of IsLegacyScalar and IsLegacyVector are also
gone; users should now get them from OpKernel.
The GraphDef version history has been moved to core/public/version.h to
minimize the number of files that must be changed in future CLs.
Change: 111947842 | Geoffrey Irving | geoffreyi@google.com | 1,452,614,816,000 |
33ee45ec434f185ca1820f494cdcd45090af9394 | Make the behavior of convert_image_dtype more explicit in the docs.
Change: 111949505 | Martin Wicke | wicke@google.com | 1,452,614,840,000 |
a7d30acd590d51f0099802835495208c5fe3f050 | Fix a minor documentation error.
Change: 111949630 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,614,848,000 |
b989c69683ba5ab5b41560575b16a6f63c803fef | Merge commit for internal changes | Manjunath Kudlur | keveman@gmail.com | 1,452,614,873,000 |
ccf01f9d77b28b649777f5a937a295f6dee2a130 | Deleted part of the local copy of Eigen since TensorFlow now pull
directly from upstream. | Benoit Steiner | benoit.steiner.goog@gmail.com | 1,452,625,066,000 |
42673c3a588ce4cc20b02ab20e5e9d38b64a3cb4 | Deleted the remainder of the local copy of eigen that is shipped with
TensorFlow. | Benoit Steiner | benoit.steiner.goog@gmail.com | 1,452,625,900,000 |
5cfb81a474169f959a430b7e51efa00621f98c95 | Merge pull request #753 from benoitsteiner/master
Delete the local copy of eigen that is shipped with tensorflow | Martin Wicke | martin.wicke@gmail.com | 1,452,626,556,000 |
e9434718c2c4f7e1de5488a411a2ea89e2724724 | Fix overly pessimistic shape inference in tf.batch_matmul for matrices with partially specified shapes.
Change: 111953111 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,630,503,000 |
36ace8a693877e8eb172bb96bdbf17c3f86d15a3 | Make Switch to always colocate with its data input. One main use of cond is conditional read of remote variables, which tries to avoid expensive remote read of a variable if certain condition is true. The graph looks like: Variable(ps) -> Switch -> SomeOp(worker). If Switch is colocated with SomeOp, Send/Recv would be inserted before Switch and always executed. We therefore colocate Switch with its data input (Variable).
Change: 111961218 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,630,514,000 |
4184fd26303416f08fb3335ae0e3edf92d634c09 | Moving the type_index implementation from resource_mgr.h to a separate file.
This enables tensorflow to be built with no rtti on all platforms if one
chooses to. Previously RTTI is only disabled with Android builds.
To build, one need to pass in explicitly two flags, using bazel:
--copt "-DGOOGLE_PROTOBUF_NO_RTTI" --cxxopt "-fno-rtti"
Change: 111963304 | Yangqing Jia | jiayq84@gmail.com | 1,452,630,525,000 |
5b68a7e8703ea58c603a0be3b6aea4e097ffbc81 | Fixes 674 from github: Use of deprecated (Py2.6) / removed (Py3) BaseException.message
Change: 111964187 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,630,537,000 |
396fdf89a736b6935ac4360ae679b1aacba14c49 | Adjust EventMgr polling for better performance.
Some recent changes to EventMgr polling that improved performance on a single-GPU benchmark had the effect of decreasing performance on a non-public distributed 4-GPU benchmark.
Those changes eliminated the immediate poll after ThenDeleteTensors, and batched together tensor deletions on the same stream into single Event groups. For the 4-GPU program, the effect was to randomly introduce longer latencies before complete Event detection.
Study suggested that the 4-GPU program with its additional GPU to GPU copy operations did not generate PollEvents calls at sufficiently high frequency, merely as a byproduct of Op queuing. The solution appears to be to increase the cycle frequency of the backup polling loop, effectively promoting it to the primary servicing mode. Contrary to prior expectations, the cost of doing so is relatively modest. In order to keep the cost even lower, in particular for mobile devices, whenever the event queue is empty, the polling loop suspends on a condition variable.
This CL lowers average step time of the 128x128, 4-GPU 6 layer block_mm benchmark from about 5 msec to about 2.4 msec (50% speedup).
It also raises the word/sec rate of ptb_word_lm from about 5500 wps to about 5600 wps.
In both cases, the additional cost of lowering the polling cycle delay from 1000 usec to 10 usec was about 0.1 CPU core. Dropping delay below 10 usec did not result in speedup to either benchmark.
Change: 111967241 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,630,547,000 |
e554a5eb8ad9c3dac43a5a61761942bed49c418b | Added control dependencies on grad to prevent ops from being computed too early.
Change: 111971225 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,630,557,000 |
fa427af8b8f580ce011b87450bbeb6fcc751b099 | Small tweak to the paths in sample notebooks.
(This was originally suggested by @freeman-lab in a copy of these notebooks.)
Change: 111971996 | Craig Citro | craigcitro@google.com | 1,452,630,571,000 |
dadbc5ce83e4df8d3c79e5e612bed25742b4c337 | Restore shape functions for SparseSegmentSqrtN accidentally reverted by cl/111953111.
Disable invalid-name pylint warning for _SparseSegmentReductionGradShape.
Change: 111977841 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,646,417,000 |
fc2e05271c537eebf371dea425ad22276feada25 | Update Bazel instructions to say 'latest stable build of bazel',
since it changes a lot.
Change: 111985892 | Vijay Vasudevan | vrv@google.com | 1,452,646,425,000 |
0684e6b3e2dba3bb838cbbddf78a046a179c441e | Adding convolution2d to tf.learn.
This follows the same basic pattern as fully_connected; the name was chosen to not directly conflict with conv2d.
Change: 111990104 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,646,435,000 |
30b8a6c1e2920ca28859b68b2a8a10d5999c4170 | Add support for device-specific graph rewriting.
This CL adds a virtual method 'MaybeRewriteGraph' to Device to enable prototyping of TensorFlow device implementations that need to modify their GraphDef before execution.
This functionality should be considered experimental and may be subject to revision in future releases.
Change: 111990489 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,646,444,000 |
1a68e5079dfb9af9f554689c0234d4bd3a1a006f | TensorFlow: fully qualify testing namespace for gunit.
Change: 111991433 | Vijay Vasudevan | vrv@google.com | 1,452,646,457,000 |
2f1ac5109dbfec33504f27ee19595e0ae7a96cce | Update README file with instructions to update and download from GCE.
Change: 111994516 | Vincent Vanhoucke | vanhoucke@google.com | 1,452,646,468,000 |
2b3395a7a997f229e19b8b3fb75bd9d1430ca2fb | Moves the typecast of a returned value to a np.array from C++ to
python, where it seems more robust in handling various python types.
Change: 111997860 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,646,477,000 |
38d30b6c59fa6110d1136a72f309b28325d1c4d3 | Move trivial Status destructor to be inline. Having it out of line
sometimes prevents other routines from being leaf routines from the
optimizer's perspective (and code size is nearly the same either way).
Change: 111998733 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,646,487,000 |
94538a944ed71eb8c0c22213fef245e09165f935 | Added OpKernel::is_internal and used it to avoid a memcmp against the
OpKernel::type_string() for every node executed on a GPU.
Change: 111998968 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,646,500,000 |
36b25e83bf022ad6b67bb22ac5adb4c0243a78a7 | Make .h includes more comprehensive in tensorflow/core/.
Change: 111999765 | Josh Levenberg | josh11b@tensorflow.org | 1,452,646,512,000 |
b18bd346f523092af2beb5c1a10a8b35a9698188 | Provide a gradient for tf.fill
fixes github issue #686
Change: 112001823 | David G. Andersen | dga@google.com | 1,452,646,521,000 |
7a524d4de0a0da527f355adb7eccea7756c82dac | Fix fill gradient test for legacy scalar strictness
Change: 112005548 | David G. Andersen | dga@google.com | 1,452,648,692,000 |
080a1f6c64976b11437d2dc4ef4178e4ca1205b7 | Merge commit for internal changes | Vijay Vasudevan | vrv@google.com | 1,452,648,788,000 |
b14b86cef14e6205a7bd4c602834008e568ea31f | Fixes with datafeeder, issue #50 at github. | makseq-ubnt | makseq@gmail.com | 1,452,691,278,000 |
5515148be85f369484d6179b7c1baab30995b068 | Minor fixes (related) to gpu_event_mgr_test.cc
This test got out of sync after some recent changes.
This CL allows the test function to explicitly stop/start
the polling loop so we can test some invarients without
non-deterministic timing issues. Also, the change
to TensorReferenceVector is now handled correctly in ~EventMgr,
and reliably tested.
Change: 112016082 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,698,652,000 |
03b5bef1ddbb5841aaf059a8b77267becbe8ea21 | Merge commit for internal changes | Manjunath Kudlur | keveman@gmail.com | 1,452,698,693,000 |
2585a3158928a82dba7e2d8875ba554e391aa9a2 | Checking if list is empty before trying to access last element. | panmari | smmuzi@gmail.com | 1,452,638,863,000 |
06feca0bf721245bd8d801b920393e4d3945aa2b | Merge pull request #756 from panmari/saver_empty_list_fix. Closes #756 | Vijay Vasudevan | vrv@google.com | 1,452,708,017,000 |
e6096fac768e6348f1d6b7c0dc475f56383c65eb | Update README.md | Yuan (Terry) Tang | terrytangyuan@users.noreply.github.com | 1,452,715,069,000 |
0c62942e2bda15fa815fff440c9f5e466bff59b5 | Change visibility to public in bower.BUILD
A remote repository cannot refer to the main repository that way and a fix in a future Bazel release will break TensorFlow with the current set-up.
Error: http://ci.bazel.io/job/TensorFlow/BAZEL_VERSION=HEAD,PLATFORM_NAME=linux-x86_64/74/console
Change: 112056797 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,722,716,000 |
9e03a7b62f75cf56766df6e22a79c6fbd1828538 | Add comparison operators for remaining realnumbertypes
uint8, int8, and int16 were missing for =, !=, <, >, <=, >=.
Change: 112066815 | Geoffrey Irving | geoffreyi@google.com | 1,452,722,727,000 |
9ac51a167cd9959a89a2b15112fcc90886fce121 | TensorFlow: remove use of bfc_allocator flag, since it defaults to true.
Change: 112067966 | Vijay Vasudevan | vrv@google.com | 1,452,722,736,000 |
9aa8a1d726c69ab60bf57e2eab1e08fdd9521d7d | Improved documentation for sigmoid_cross_entropy_with_logits.
The definition of logistic loss used by sigmoid_cross_entropy_with_logits is a
bit unusual. It took me some time to convince myself that it was equivalent to
a more common definition of logistic loss. I'd like to document the equivalence
because I think it will be helpful for other TensorFlow newbies.
Change: 112075662 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,722,747,000 |
0b1db1ca09d858b526699f0c0578c7de8f6623b5 | Merge commit for internal changes | Vijay Vasudevan | vrv@google.com | 1,452,723,288,000 |
ec98c714350f5e7cd82177a79c19345ad14cfd81 | Fix typos in tutorials | Dongjoon Hyun | dongjoon@apache.org | 1,452,729,314,000 |
27c7dd72bedee9e46de43c8a4d164513c4bf5924 | Merge pull request #765 from dongjoon-hyun/fix_typos_in_tutorials. Closes #765 | Vijay Vasudevan | vrv@google.com | 1,452,729,909,000 |
ee40b6e6e646eb31adf1867fda370380a7dc74e8 | Post order = Reverse topological order
Change: 112091177 | Manjunath Kudlur | keveman@gmail.com | 1,452,734,980,000 |
cbdf278dbd998aab57898763605f6863e9b42b76 | Added roadmap.
Fixes #162.
Change: 112102777 | Martin Wicke | wicke@google.com | 1,452,734,998,000 |
6dbfb95100b73ad26ebebb9be9c0429dc0cece8a | Avoid allocation of std::function<> object per node executed, as
well as a hash-table lookup per allocated output.
Instead, we now pre-compute the AllocatorAttributes for every output
tensor in the graph into an array (indexed by a base number per node +
output index), and changed OpKernelContext::Params to provide
a pointer to the base of the array for the node, rather than providing
a std::function<>.
Updated test code to avoid so much code duplication when setting up
the OpKernelContext::Params object in various places.
Used gtl::InlinedVector<...> instead of std::vector<...> in a few
places in tensorflow/core/kernels/reduction_ops_common.h
Didn't make a measurable change in overall performance but allocations and
time spent in the std::function destructor code was significantly reduced.
Change: 112103260 | A. Unique TensorFlower | nobody@tensorflow.org | 1,452,735,008,000 |
8e388a5546f1466aa0d2afa00e5a015997a23a2b | TensorFlow: Get rid of legacy command line flags use in TensorFlow.
Change: 112105282 | Vijay Vasudevan | vrv@google.com | 1,452,735,015,000 |
a2c5e74b3aa02fb3061e040a62ccf905e2b051f2 | Merge commit for internal changes | Vijay Vasudevan | vrv@google.com | 1,452,735,060,000 |
a2e30a5ddec06273072ad5e7de331caaa22a6676 | + add_to_index for dask_io | terrytangyuan | terrytangyuan@gmail.com | 1,452,735,451,000 |
a4fd70f9bccd420250d700e0e707679c43bcf763 | + _get_divisions() for dask_io | terrytangyuan | terrytangyuan@gmail.com | 1,452,735,695,000 |
7b8ff3df8d0e376eb986acba5933673e159c7144 | + check for dd.dataframe | terrytangyuan | terrytangyuan@gmail.com | 1,452,735,844,000 |
6314d032c263a54ff4b1e2b8959db5ee796bcb0e | + _construct_dask_df_with_divisions for dask_io | terrytangyuan | terrytangyuan@gmail.com | 1,452,736,064,000 |
d1576117f8fda62ae9108fca7e363ee88ed3f667 | * Fixed dask data extractors | terrytangyuan | terrytangyuan@gmail.com | 1,452,736,894,000 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.