profile
viewpoint

IBM/adversarial-robustness-toolbox 1505

Python library for adversarial machine learning (evasion, extraction, poisoning, verification, certification) with attacks and defences for neural networks, logistic regression, decision trees, SVM, gradient boosted trees, Gaussian processes and more with multiple framework support

tensorflow/addons 864

Useful extra functionality for TensorFlow 2.x maintained by SIG-addons

twosixlabs/armory 24

ARMORY Adversarial Robustness Evaluation Test Bed

seanpmorgan/tf-weightnorm 12

Examples for TensorFlow Weight Normalization

seanpmorgan/tensorflow 1

Computation using data flow graphs for scalable machine learning

dynamicwebpaige/community-1 0

Stores documents used by the TensorFlow developer community

seanpmorgan/addons 0

Useful extra functionality for TensorFlow maintained by SIG-addons

seanpmorgan/adversarial-robustness-toolbox 0

Python library for adversarial machine learning, attacks and defences for neural networks, logistic regression, decision trees, SVM, gradient boosted trees, Gaussian processes and more with multiple framework support

seanpmorgan/armory 0

ARMORY Adversarial Robustness Evaluation Test Bed

seanpmorgan/armory-example 0

Example external repository for interacting with armory.

delete branch seanpmorgan/armory

delete branch : refactor-vm-space

delete time in 14 hours

PR opened twosixlabs/armory

Reviewers
Refactor clearing space on VM
+4 -10

0 comment

2 changed files

pr created time in a day

create barnchseanpmorgan/armory

branch : refactor-vm-space

created branch time in a day

startedcommaai/openpilot

started time in 2 days

push eventseanpmorgan/armory

ng390

commit sha d6f2ed983cb4da0803585b37373311e77bb3fd8d

Minor documentation updates (#600) * Minor typos * Update path reference in docs

view details

davidslater

commit sha 9d9868d969cbe2cad91f3016c384e79493abab06

PGD fix (also applies to other attacks that need ground truth labels) (#597) * update test * updated baselines and scenarios * minor update * update to use labels * update to use labels * Update documentation and schema * Update docs/configuration_files.md Co-authored-by: ng390 <neal.gupta@twosixlabs.com> * Don't change official scenarios Co-authored-by: Sean Morgan <seanmorgan91@gmail.com> Co-authored-by: ng390 <neal.gupta@twosixlabs.com>

view details

ng390

commit sha c0b7c4fea168e0e058f250da720a910519cfb058

More user friendly and configurable output filename/directory (#594) * Configurable output filename/directory, revert uuid4 to timestamp * Update docs * Update configs * Update config schema

view details

ng390

commit sha 76bf4e0457b90736d57c5ea42ab7c5a0e0e77101

Dataset split and preprocessing docs (#599) * Update docs * Add reference to digits dataset split, typos * Clarification of splits for resisc45 * Update spacing * Update * Update * Update Co-authored-by: Sean Morgan <seanmorgan91@gmail.com>

view details

Sean Morgan

commit sha 2ad5c084b7f4b3a355049e2ad79148b8d31dd179

Clear space on VM for builds (#603)

view details

Sean Morgan

commit sha ffe4ea4799243b1e34417c6fc399a212f3eaa5b9

Update to 0.9 (#602)

view details

push time in 2 days

startedfacebookresearch/detr

started time in 3 days

PR opened twosixlabs/armory

Clear space on VM for builds
+10 -0

0 comment

2 changed files

pr created time in 4 days

PR opened twosixlabs/armory

Reviewers
Clear space on VM for builds
+10 -0

0 comment

2 changed files

pr created time in 4 days

create barnchseanpmorgan/armory

branch : clear-space

created branch time in 4 days

PR opened twosixlabs/armory-example

Reviewers
Update official scenarios
+91 -27

0 comment

13 changed files

pr created time in 4 days

create barnchseanpmorgan/armory-example

branch : update-verison

created branch time in 4 days

push eventseanpmorgan/armory-example

superctj

commit sha 5fe7f9c97c7182a60abcd4c6f3b254d3bbab9830

Add tf1 example

view details

superctj

commit sha 876f0bbf130886861233089b3af51b0d9059a6f0

Update style

view details

Sean Morgan

commit sha 7655693a7bbe5fc7d4742443aa2921640564d204

Cleanup used budget from configuration files (#43)

view details

superctj

commit sha 59f2872ba6986f81d6526ca4d847557350767ddb

Commit requested changes

view details

Sean Morgan

commit sha 6bf908bdce1bae4d56c353a1bee225e978cd7cc6

Update example_scenario_configs/cifar10_pgd_tf1.json

view details

Sean Morgan

commit sha 3b1924a9f93049ef537aec558280ffd4cf7b8951

Update example_scenario_configs/cifar10_pgd_tf1.json

view details

Sean Morgan

commit sha 2546460ad2f5ce088e34790a664771f2bb895398

Update naming

view details

Sean Morgan

commit sha 4b13a08d8ca17c64e160ba14a5b178a4f8f24f0a

Merge pull request #42 from superctj/tf-example Add tf1 example

view details

push time in 4 days

PR opened twosixlabs/armory

Reviewers
Update to 0.9
+23 -23

0 comment

23 changed files

pr created time in 4 days

create barnchseanpmorgan/armory

branch : bump-0.9

created branch time in 4 days

push eventtwosixlabs/armory

ng390

commit sha d0679bd5b7d0308617ba580fde7cd65f1841e841

Update docker container names (#601) * Update configs * Remove extraneous file

view details

push time in 4 days

PR merged twosixlabs/armory

Update docker container names
+22 -22

0 comment

22 changed files

ng390

pr closed time in 4 days

push eventtwosixlabs/armory

ng390

commit sha 76bf4e0457b90736d57c5ea42ab7c5a0e0e77101

Dataset split and preprocessing docs (#599) * Update docs * Add reference to digits dataset split, typos * Clarification of splits for resisc45 * Update spacing * Update * Update * Update Co-authored-by: Sean Morgan <seanmorgan91@gmail.com>

view details

push time in 4 days

PR merged twosixlabs/armory

Dataset split and preprocessing docs

Fixes #515

+51 -17

0 comment

1 changed file

ng390

pr closed time in 4 days

issue closedtwosixlabs/armory

Documentation for the armory supported datasets

Can we please have some documentation about the splits, code used to generate the splits and any pre-processing done on the cached datasets? For example, the 'validation' split is present for resisc45 datasets but is absent in mnist and cifar10. It would be nice to have consistent dataset splits.

closed time in 4 days

joemathai

push eventng390/armory

Sean Morgan

commit sha 492e940ab7cc94aa4fac1c78af71e310215f0bf4

Update

view details

push time in 4 days

push eventng390/armory

Sean Morgan

commit sha 3d36a8ddf45d226df2ac67913870d0f0a37b8ea4

Update

view details

push time in 4 days

push eventng390/armory

Sean Morgan

commit sha 715d980b781b2da1de56e8464c93b9165034f03b

Update

view details

push time in 4 days

push eventng390/armory

Sean Morgan

commit sha 9727aeddad313cc37cb3b0ca3c9db6a041ab99f4

Update spacing

view details

push time in 4 days

push eventtwosixlabs/armory

ng390

commit sha c0b7c4fea168e0e058f250da720a910519cfb058

More user friendly and configurable output filename/directory (#594) * Configurable output filename/directory, revert uuid4 to timestamp * Update docs * Update configs * Update config schema

view details

push time in 4 days

PR merged twosixlabs/armory

More user friendly and configurable output filename/directory

Fixes #584. Draft (1) because wanted to discuss timestamp in directory and filename (2) need to update docs.

+81 -3

0 comment

26 changed files

ng390

pr closed time in 4 days

issue closedtwosixlabs/armory

Make output files more user friendly or configurable

Right now the output directory is a random number, once you run many experiments finding which result file corresponds to which experiment is not going to be easy.

closed time in 4 days

davidslater

push eventtwosixlabs/armory

davidslater

commit sha 9d9868d969cbe2cad91f3016c384e79493abab06

PGD fix (also applies to other attacks that need ground truth labels) (#597) * update test * updated baselines and scenarios * minor update * update to use labels * update to use labels * Update documentation and schema * Update docs/configuration_files.md Co-authored-by: ng390 <neal.gupta@twosixlabs.com> * Don't change official scenarios Co-authored-by: Sean Morgan <seanmorgan91@gmail.com> Co-authored-by: ng390 <neal.gupta@twosixlabs.com>

view details

push time in 4 days

issue closedtwosixlabs/armory

PGD attack generates incorrect values

This is due to a couple of issues: attack generation does not take the ground truth value into account, and the attack metrics are computed regardless of whether the initial classification was correct.

The fix should make use of ART's full attack.generate functionality

closed time in 4 days

davidslater

push eventtwosixlabs/armory

ng390

commit sha d6f2ed983cb4da0803585b37373311e77bb3fd8d

Minor documentation updates (#600) * Minor typos * Update path reference in docs

view details

push time in 4 days

PR merged twosixlabs/armory

Minor documentation updates

Minor documentation updates

+16 -13

0 comment

6 changed files

ng390

pr closed time in 4 days

push eventdavidslater/armory

Sean Morgan

commit sha f9f1771cd2f7e48be9230afb36ef2c85a2dfb054

Don't change official scenarios

view details

Sean Morgan

commit sha a555ac8470fe1b4e22552b28dc5ab7937c3ed78e

Merge remote-tracking branch 'davidslater/pgd-attack' into pr/davidslater/597

view details

push time in 5 days

push eventdavidslater/armory

Sean Morgan

commit sha 6f62625e225222df198865e9a153e045af3ab62f

Update docs/configuration_files.md Co-authored-by: ng390 <neal.gupta@twosixlabs.com>

view details

push time in 5 days

push eventdavidslater/armory

Sean Morgan

commit sha 8ac718d21f9aa65cdf6e5c1bdd47fa20dc58a70e

Update documentation and schema

view details

push time in 5 days

push eventseanpmorgan/armory

grobertson-ext

commit sha 7f6f2ae99d5dedf519f66b9e9249921acc646283

RESISC45 attacks in png format (#568) * fix accuracy calculation * add image densenet/resisc scenario * fix accuracy calculation for example resisc scenario -- affects output to Evaluation Results written <output_dir>/evaluation-results.json * run black on accuracy fix * add densenet scenario * correct scenario file in json, and update total_iterations to batches_per_epoch * image FGM attack scenario with densenet model and resisc45 dataset * black resisc45_scenario * For DenseNet image scenario, if not loading pretrained weights, use Imagenet pre-trained weights * add dataset generator file in armory/data * add rest of files without checksum complete * Adversarial patch attack for image classification * add univ patch and univ pert for img resisc scenario in phase 1 * resisc universal patch and universal perturbation attacks * resisc universal patch and universal perturbation attacks * uploaded file to s3 * moved to adversarial datasets * remerge master * remerge master * adversarial resisc config * v2 (png) of the resisc adversarial dataset * change version number to 1.0.1, remove _v2 from dataset name * updated to use s3 * updated s3 cache Co-authored-by: David Slater <david.slater@twosixlabs.com>

view details

yusong-tan

commit sha 027155c3721f9c4ad57fc33ef8472327dc4e8175

Adversarial datasets docs (#596) * init commit * more edits * typo * typo * added hyperlinks between datasets.md and adversarial_datasets.md * typo * Update adversarial_datasets.md Co-authored-by: davidslater <david.slater@twosixlabs.com>

view details

Sean Morgan

commit sha 3b2d6275885ee4d6b579a4005dee6b148dd981e1

* Increase default SHM size (#595)

view details

push time in 5 days

issue commenttensorflow/addons

Keypoints class/ops

Thanks for the update... please keep us posted with information/timelines as they become available. In either case we will not be providing this functionality in Addons as it would be duplicated.

Thanks @seanpmorgan @bhack collaborating on this! Meanwhile it'd be great to re-direct them to the new repo so we could accept contribution as well :-)

Yes, for those that end up on this issue please check out the following repository and/or submit contributions there: https://github.com/keras-team/keras-cv

bhack

comment created time in 5 days

issue closedtensorflow/addons

Implement fast-nms

Describe the feature and the current behavior/state. The fast-nms algorithm has better performance than traditional nms. Described in paper. Kernel implementation may provides better performance than python implementation. Relevant information

  • Are you willing to contribute it (yes/no): yes
  • Are you willing to maintain it going forward? (yes/no): yes
  • Is there a relevant academic paper? (if so, where): https://arxiv.org/pdf/1904.02689.pdf
  • Is there already an implementation in another framework? (if so, where): https://github.com/dbolya/yolact/blob/1722387d75210361c1f21c911d0e2420a48c7a23/layers/functions/detection.py#L136
  • Was it part of tf.contrib? (if so, where): no

Which API type would this fall under (layer, metric, optimizer, etc.)

Who will benefit with this feature?

Any other info. Tensorflow Implementation on colab

closed time in 5 days

fsx950223

issue commenttensorflow/addons

Build GPU custom Ops for windows

Given the recent news: https://developer.nvidia.com/cuda/wsl Is this feature still requested from the community?

seanpmorgan

comment created time in 5 days

issue closedtensorflow/addons

Use isort in the pre-commit hook

https://github.com/timothycrosley/isort

after the 5.0 release, it will be compatible with black. Do we want to use it?

closed time in 5 days

gabrieldemarmiesse

issue commenttensorflow/addons

Use isort in the pre-commit hook

Thinking on this a little more, I'm not sure the workload for reviewing all of these PRs would be worth the effort. Happy to reconsider but closing for now.

gabrieldemarmiesse

comment created time in 5 days

issue closedtensorflow/addons

[2/2] Add Image Ops for Data Augmentation

Part 1: https://github.com/tensorflow/addons/issues/1333

As part of #1226, relevant image ops need to be added to tfa.image. List of ops tracked in this issue (tick the op only once the corresponding PR is merged):

  • [x] to grayscale
  • [ ] color jitter
  • [ ] color jitter nonrand
  • [ ] color jitter rand
  • [ ] compute crop shape
  • [ ] center crop
  • [ ] distorted bounding box crop
  • [ ] crop and resize
  • [x] gaussian blur
  • [ ] random crop with resize
  • [ ] random color jitter
  • [ ] random blur
  • [ ] batch random blur

Reference: https://github.com/google-research/simclr/blob/master/data_util.py

closed time in 5 days

vinayvr11

issue commenttensorflow/addons

[2/2] Add Image Ops for Data Augmentation

Per https://github.com/tensorflow/addons/issues/1226#issuecomment-634755922 the autoaugment/randaugment features will be a part of https://github.com/keras-team/keras-cv

Apologies for not being able to present this information sooner to the contributors.

vinayvr11

comment created time in 5 days

issue closedtensorflow/addons

[1/2] Add Image Ops for Data Augmentation

Part 2: https://github.com/tensorflow/addons/issues/1353

As part of #1226, relevant image ops need to be added to tfa.image. List of ops tracked in this issue (tick the op only once the corresponding PR is merged):

  • [x] Blend
  • [x] CutOut
  • [ ] Identity
  • [x] Translate X, Y
  • [x] Shear X, Y
  • [ ] Solarize Ops
  • [ ] Posterize
  • [ ] AutoContrast
  • [x] Equalize
  • [ ] Invert
  • [ ] Color
  • [ ] Brightness
  • [x] Sharpness

Reference: https://github.com/tensorflow/tpu/blob/master/models/official/efficientnet/autoaugment.py

closed time in 5 days

abhichou4

issue commenttensorflow/addons

[1/2] Add Image Ops for Data Augmentation

Per https://github.com/tensorflow/addons/issues/1226#issuecomment-634755922 the autoaugment/randaugment features will be a part of https://github.com/keras-team/keras-cv

Apologies for not being able to present this information sooner to the contributors.

abhichou4

comment created time in 5 days

issue closedtensorflow/addons

Add Bounding Box Ops

As part of #1226, relevant image bounding box operations need to be added to tfa.image. List of ops tracked in this issue (tick the op only once the corresponding PR is merged):

  • [ ] TranslateX BBox
  • [ ] TranslateY Only BBoxes
  • [ ] ShearX BBox
  • [ ] Rotate BBox
  • [ ] Cutout Only BBoxes
  • [ ] ShearY BBox
  • [ ] ShearY Only BBoxes

Reference: https://github.com/google-research/simclr/blob/master/data_util.py

closed time in 5 days

vinayvr11

issue commenttensorflow/addons

Add Bounding Box Ops

Per https://github.com/tensorflow/addons/issues/1366#issuecomment-634902190 these features would better fit in: https://github.com/keras-team/keras-cv

vinayvr11

comment created time in 5 days

Pull request review commenttensorflow/addons

Fix MultiHeadAttention docs

 class MultiHeadAttention(tf.keras.layers.Layer):         bias_regularizer: regularizer, regularizer for the bias weights.         bias_constraint: constraint, constraint for the bias weights. -    Call Arguments-        inputs:  List of the following tensors:-            * `query`: Tensor of shape `(..., query_elements, query_depth)`-            * `key`: `Tensor of shape '(..., key_elements, key_depth)`-            * `value`: Tensor of shape `(..., key_elements, value_depth)` (optional)+    Call Arguments:+        inputs:  List of the tensors `[query, key, value]` where  `query` has shape `(..., query_elements, query_depth)`, `key` has shape '(..., key_elements, key_depth)`, and `value` has shape `(..., key_elements, value_depth)`. `value` is optional, if not given `key` will be used.

Yes I would prefer we stick to what's being done elsewhere in the ecosystem. I think if we copy the formatting as linked above it comes out quite readable in the generated docs. What do you think @cgarciae ?

cgarciae

comment created time in 5 days

issue closedtensorflow/addons

Migrate AutoAugment and RandAugment to TensorFlow Addons.

Describe the feature and the current behavior/state. RandAugment and AutoAugment are both policies for enhanced image preprocessing that are included in EfficientNet, but are still using tf.contrib.

https://github.com/tensorflow/tpu/blob/master/models/official/efficientnet/autoaugment.py

The only tf.contrib image operations that they use, however, are rotate, translate and transform - all of which have been included in TensorFlow Addons.

Relevant information

  • Are you willing to contribute it (yes/no): No, but am hoping that someone from the community will pick it up (potentially a Google Summer of Code student)?

  • Are you willing to maintain it going forward? (yes/no): Yes

  • Is there a relevant academic paper? (if so, where): AutoAugment Reference: https://arxiv.org/abs/1805.09501 RandAugment Reference: https://arxiv.org/abs/1909.13719

  • Is there already an implementation in another framework? (if so, where): See link above; this would be a standard migration from tf.contrib.

  • Was it part of tf.contrib? (if so, where): Yes

Which API type would this fall under (layer, metric, optimizer, etc.) Image

Who will benefit with this feature? Anyone doing image preprocessing, especially for EfficientNet.

closed time in 5 days

dynamicwebpaige

issue commenttensorflow/addons

Migrate AutoAugment and RandAugment to TensorFlow Addons.

Closing as this is now going to be implemented elsewhere in the ecosystem. Since we've already merged a few components of these we can deprecate them as they're available.

dynamicwebpaige

comment created time in 5 days

issue closedtensorflow/addons

Keypoints class/ops

Describe the feature and the current behavior/state. Add keypoint class utils Relevant information

  • Are you willing to contribute it (yes/no): no
  • Are you willing to maintain it going forward? (yes/no): probably
  • Is there a relevant academic paper? (if so, where): Yes but it is a too popular topic to point any specific paper.
  • Is there already an implementation in another framework? (if so, where): Yes in official Facebook Pytorch detectron2 https://github.com/facebookresearch/detectron2/blob/master/detectron2/structures/keypoints.py
  • Was it part of tf.contrib? (if so, where): No we had some sparse keypoints utils/ops in model https://github.com/tensorflow/models/blob/master/research/object_detection/predictors/heads/keypoint_head.py https://github.com/tensorflow/models/blob/master/research/object_detection/core/keypoint_ops.py Which API type would this fall under (layer, metric, optimizer, etc.) keypoints Who will benefit with this feature? Many applications like human pose, hande pose, face landmarks, 3d objects cuboid etc.. Any other info. In Mediapipe there are many pre/postprocessing operations over keypoints that rely on Opencv keypoints with the Tensorflow lite inference. Probably some of these operations could put in graph if we can handle keypoints in Tensorflow directly.

closed time in 5 days

bhack

issue commenttensorflow/addons

Keypoints class/ops

Thanks for the update... please keep us posted with information/timelines as they become available. In either case we will not be providing this functionality in Addons as it would be duplicated.

bhack

comment created time in 5 days

push eventseanpmorgan/community

Sean Morgan

commit sha 8f1af584c6ce7f6ba6edc42e28413b50420ccfee

* Update transition

view details

push time in 5 days

push eventseanpmorgan/addons

Sean Morgan

commit sha f85e2af64dd3674f4e802133672c02ea32986216

* Update dockerfile dependencies

view details

Sean Morgan

commit sha 8b0f05e114f9e9281ff287e7232a0225df9e8c39

* Modularize build script

view details

push time in 5 days

push eventseanpmorgan/community

Sean Morgan

commit sha a3b8b6d8547c30ed1a6d51f3b6a4085cd43ea885

Spacing

view details

push time in 5 days

push eventseanpmorgan/community

Sean Morgan

commit sha 8fc2dd35cf0aa291c96ff519a5110bb6310b9f80

typos

view details

push time in 5 days

push eventseanpmorgan/community

Sean Morgan

commit sha 6a8f78967c37c5ebf27d5105233ccd99b25cd6bf

* gelu migration RFC

view details

push time in 5 days

delete branch seanpmorgan/armory

delete branch : increase-shm-size

delete time in 5 days

issue openedtwosixlabs/armory

Pre-defined poisoned data splits

Per discussions, it'd be a good idea to evaluate poisoning defenses against several pre-defined poisoned data splits. This ensures robustness against different initializations of poisoned samples and also keeps the evaluation consistent for various defenses. This would be in contrast to the current random.choice of images to poison.

created time in 5 days

PR opened twosixlabs/armory

Reviewers
Increase default SHM size

So went through a couple of iterations on deciding if we wanted to introduce docker configs to the scenario configurations. Ideally, we don't since they require special parsing from ECS task conversion. Looking into /dev/shm and the tempfs spec it doesn't appear to have negative consequences to hard code this as a higher value: https://unix.stackexchange.com/questions/259823/if-tmpfs-has-bigger-size-than-your-ram-how-much-ram-does-it-use-are-applicatio

I tried setting this to 64GB (2x my laptops memory) and everything worked fine so shouldn't be an issue for systems < 16GB RAM. This guide also recommends increasing shm by default: https://www.pugetsystems.com/labs/hpc/How-To-Setup-NVIDIA-Docker-and-NGC-Registry-on-your-Workstation---Part-5-Docker-Performance-and-Resource-Tuning-1119/

@adamj26 should this get merged we'll need to hardcode this into the ECS task conversion.

Closes #535

+1 -0

0 comment

1 changed file

pr created time in 5 days

create barnchseanpmorgan/armory

branch : increase-shm-size

created branch time in 5 days

Pull request review commenttensorflow/addons

Fix MultiHeadAttention docs

 @tf.keras.utils.register_keras_serializable(package="Addons") class MultiHeadAttention(tf.keras.layers.Layer):     r"""-    MultiHead Attention layer.+    MultiHead Attentsion layer.
    MultiHead Attention layer.
cgarciae

comment created time in 5 days

issue commenttensorflow/addons

tensorflow.python.framework.errors_impl.NotFoundError: Op type not registered 'GatherTree' in binary running

Hmmm TFA 0.11 hasn't yet been released are you sure you're correct about the versions. FWIW TFA 0.10 requires TensorFlow >= 2.2 which could be causing this error.

gyshi

comment created time in 6 days

push eventseanpmorgan/armory

yusong-tan

commit sha 2c81952c7eee232d249fb08426d25ae02f722407

Added baseline performance for image scenario (#589) * Added baseline performance for image scenario * edited scenarios.md to reference correct art_experimental defences for image and video scenarios * fixed typo

view details

Sean Morgan

commit sha d03a537397d0aa938197adc56d3624fb6165a588

Fix port mapping for armory launch and armory run (#593) * * Fix port mapping for armory launch and armory run * typo

view details

push time in 6 days

delete branch seanpmorgan/armory

delete branch : fix-port-mapping

delete time in 6 days

issue closedtensorflow/addons

code does not match doc for extending optimizer with weight decay

System information

  • OS Platform and Distribution (e.g., Linux Ubuntu 16.04): n/a
  • TensorFlow version and how it was installed (source or binary): current in github
  • TensorFlow-Addons version and how it was installed (source or binary): n/a
  • Python version: n/a
  • Is GPU used? (yes/no): n/a

Describe the bug

The docs for decoupling weight decay extension do not match the library.

Code to reproduce the issue

n/a

Other info / logs

In the docs (https://www.tensorflow.org/addons/api_docs/python/tfa/optimizers/weight_decay_optimizers/DecoupledWeightDecayExtension) we see that an example is given to extend an existing optimizer.

"extend_with_decoupled_weight_decay( OptimizerX, weight_decay=weight_decay)"

How in the code the signature for this method is "def extend_with_decoupled_weight_decay(base_optimizer):"

There is no keyword argument for weight_decay

closed time in 6 days

ben-arnao

issue commenttensorflow/addons

code does not match doc for extending optimizer with weight decay

Hi @yil532 it was fixed with #1718 .. thanks for bringing this up though we can close this issue.

ben-arnao

comment created time in 6 days

pull request commenttensorflow/addons

Publish dev container

Could you add

sudo update-alternatives --install /usr/bin/clang++ clang++ /usr/bin/clang++-8 100
sudo update-alternatives --install /usr/bin/clang clang /usr/bin/clang-8 100

Sure, but could you expand on why its needed. All of our documentation uses gcc and nvcc to compile. I know there was work on getting TF to compile using cuda-clang but not sure if we want to encourage that?

seanpmorgan

comment created time in 6 days

Pull request review commenttensorflow/addons

Publish dev container

+#!/usr/bin/env bash++set -x -e++df -h+docker info+# to get more disk space+rm -rf /usr/share/dotnet &

Yeah which is why I moved it into .github/workflows it is not meant to be built locally. We could make it so it calls a script in tools/ but then the user would need download the base image which is almost as large so I don't see a lot of benefit

seanpmorgan

comment created time in 6 days

pull request commenttensorflow/addons

Publish dev container

I am looking at the compressed size of devel image TF vs TFA:

https://hub.docker.com/r/tensorflow/tensorflow/tags https://hub.docker.com/r/tfaddons/tensorflow/tags

The TF CPU devel is 1/3 of our devel image and we have 1GB of compressed overhead over tensorflow GPU devel (uncompressed our is 9.2GB).

I think it could be useful to have a default small size 1GB image to lower the download overhead but I think that is still not possible to handle multiple configuration (GPU/CPU) in the dev-containers microsoft/vscode-remote-release#2143 (comment).

As discussed offline this is a consequence of https://github.com/tensorflow/tensorflow/issues/38352 . There is not much we can do without deviating from the custom-op recommended images.

seanpmorgan

comment created time in 6 days

Pull request review commentmicrosoft/vscode-dev-containers

Tensorflow addons

+#-------------------------------------------------------------------------------------------------------------+# Copyright (c) Microsoft Corporation. All rights reserved.+# Licensed under the MIT License. See https://go.microsoft.com/fwlink/?linkid=2090316 for license information.+#-------------------------------------------------------------------------------------------------------------++FROM tfaddons/dev_container:latest++# This Dockerfile adds a non-root user with sudo access. Use the "remoteUser"+# property in devcontainer.json to use it. On Linux, the container user's GID/UIDs+# will be updated to match your local UID/GID (when using the dockerFile property).+# See https://aka.ms/vscode-remote/containers/non-root-user for details.+ARG USERNAME=vscode+ARG USER_UID=1000+ARG USER_GID=$USER_UID++# Options for common package install script - SHA updated on release+ARG INSTALL_ZSH="false"+ARG UPGRADE_PACKAGES="true"

I'd prefer they not be updated in this Dockerfile but rather done in tfaddons/dev_container if needed

bhack

comment created time in 6 days

push eventtensorflow/addons

Guillaume Klein

commit sha 0b44d6ef88479c8c4a3d927bf6018ef0e5adf75d

Update shape usage to TensorFlow 2 style (#1891) * Prefer `shape` attribute over `get_shape` method * Remove usage of `tf.compat.dimension_value` (we only support TF 2) * Remove usage of deprecated attribute `TensorShape.dims`

view details

push time in 6 days

PR merged tensorflow/addons

Update shape usage to TensorFlow 2 style cla: yes seq2seq
  • Prefer shape attribute over get_shape method
  • Remove usage of tf.compat.dimension_value (we only support TF 2)
  • Remove usage of deprecated attribute TensorShape.dims
+93 -117

0 comment

9 changed files

guillaumekln

pr closed time in 6 days

PR opened twosixlabs/armory

Reviewers
Fix port mapping for armory launch and armory run

Closes #513

+24 -12

0 comment

2 changed files

pr created time in 6 days

push eventseanpmorgan/armory

Sean Morgan

commit sha 2a689ba481724aac354f93fea2a62a9ae09f6947

typo

view details

push time in 6 days

create barnchseanpmorgan/armory

branch : fix-port-mapping

created branch time in 6 days

push eventseanpmorgan/armory

davidslater

commit sha 60d0b03dda0c772ef0396b060a69a3f3b7465e7e

--check mode for armory run (#590) * --check mode for armory run * update config and run name * evaluation check-run Co-authored-by: Sean Morgan <seanmorgan91@gmail.com>

view details

push time in 6 days

push eventtensorflow/addons

Guillaume Klein

commit sha 3ac4e6bd048c3e6e79102ff65ff24f5a7ffdb9ec

Clarify cell argument docstring in decoder objects (#1890) Also remove some `TypeError` notices that are redundant with the type annotations.

view details

push time in 6 days

PR merged tensorflow/addons

Clarify cell argument docstring in decoder objects cla: yes seq2seq

Also remove some TypeError notices that are redundant with the type annotations.

+6 -15

0 comment

2 changed files

guillaumekln

pr closed time in 6 days

push eventtensorflow/addons

Guillaume Klein

commit sha bec20003693a790b829eb6fcddb809fb115fc66e

Remove workaround for TensorFlow < 2.2 (#1889)

view details

push time in 6 days

PR merged tensorflow/addons

Reviewers
Remove workaround for TensorFlow < 2.2 cla: yes seq2seq

We now target TensorFlow >= 2.2 so we can remove this workaround.

+0 -12

0 comment

1 changed file

guillaumekln

pr closed time in 6 days

push eventtwosixlabs/armory

davidslater

commit sha 60d0b03dda0c772ef0396b060a69a3f3b7465e7e

--check mode for armory run (#590) * --check mode for armory run * update config and run name * evaluation check-run Co-authored-by: Sean Morgan <seanmorgan91@gmail.com>

view details

push time in 7 days

PR merged twosixlabs/armory

--check mode for armory run

Fixes #382

Can now quickly test code in a scenario by using the CLI flag --check. This works by essentially setting all loaded datasets to have a total size equal to a single batch (note that this means that if the batch size is larger, it will take longer). This required skipping the filtering defense portion of the poisoning scenario, but otherwise works fine on all scenarios.

To test:

python -m armory scenario_configs/mnist_baseline.json --check
python -m armory scenario_configs/mnist_baseline.json --check --no-docker

I also ensure that the --debug flag is carried into the scenario from the calling function, which partially addresses #500 .

+97 -13

0 comment

6 changed files

davidslater

pr closed time in 7 days

push eventdavidslater/armory

ng390

commit sha 1f98ff21612193c86f68e28a5944a579a88f506c

Imagenet adversarial paired samples (#576) * Update imagenet-adversarial to use new format * Simplify logic * Update cache

view details

yusong-tan

commit sha 11b9aac8143901edad2919eb2f56b6bd84050c8f

Added art_experiemental JpegCompression wrapper to work with normaliz… (#588) * Added art_experiemental JpegCompression wrapper to work with normalized inputs * Update resisc45_baseline_densenet121_adversarial.json * Update jpeg_compression_normalized.py * Update resisc45_baseline_densenet121_adversarial.json Use full mean and std values from densenet121 preprocessing Co-authored-by: davidslater <david.slater@twosixlabs.com>

view details

Adam Jacobson

commit sha b20b38b48066454c12331ff7fa39dba69dc1b843

Optionally send evaluation results to a MongoDB instance (#581) * Send results to a MongoDB instance if the `--mongo` flag is given to the scenario * Fix typo in `setup.py` * Provide examples of specifying the host in the Scenario `--mongo` help text * Provide further example of connecting to MongoDB with a username and password

view details

ng390

commit sha 8b034a150cd5e0893d6901444e18a702bee33970

Poisoning scenario: update randomization (#586) * Poisoning scenario: poison target class uniformly at random rather than all target class samples in randomly chosen batches * Update poisoning randomization * Warning if no poisons generated

view details

davidslater

commit sha cd6b0b201cc0fbcd7c3fa4e4b38411ee1af5d070

SNR perturbation metrics (#573) * Add snr * added tests

view details

Sean Morgan

commit sha d396303dee3db7a59fd2c3d9914a2978fb88b84e

Merge branch 'master' into pr/davidslater/590 # Conflicts: # armory/scenarios/base.py

view details

push time in 7 days

push eventseanpmorgan/armory

Adam Jacobson

commit sha b20b38b48066454c12331ff7fa39dba69dc1b843

Optionally send evaluation results to a MongoDB instance (#581) * Send results to a MongoDB instance if the `--mongo` flag is given to the scenario * Fix typo in `setup.py` * Provide examples of specifying the host in the Scenario `--mongo` help text * Provide further example of connecting to MongoDB with a username and password

view details

ng390

commit sha 8b034a150cd5e0893d6901444e18a702bee33970

Poisoning scenario: update randomization (#586) * Poisoning scenario: poison target class uniformly at random rather than all target class samples in randomly chosen batches * Update poisoning randomization * Warning if no poisons generated

view details

davidslater

commit sha cd6b0b201cc0fbcd7c3fa4e4b38411ee1af5d070

SNR perturbation metrics (#573) * Add snr * added tests

view details

push time in 7 days

IssuesEvent

issue closedtensorflow/addons

Vectors not getting parsed on TensorFlow Projector (TripletSemiHardLoss tutorial)

Hi,

I am referring to this tutorial on TripletSemiHardLoss: https://www.tensorflow.org/addons/tutorials/losses_triplet. After I download the vecs.tsv and meta.tsv files and head over to https://projector.tensorflow.org/ and try uploading the vecs.tsv file I am getting the following:

image

It's been more than 30 minutes, my internet connection is fine. Is it possible to shed some light as to why it might happen? I understand that there might be some erroneous/corrupted entries while preparing the vecs.tsvfile but I don't think that's the case here.

closed time in 7 days

sayakpaul

issue commenttensorflow/addons

Vectors not getting parsed on TensorFlow Projector (TripletSemiHardLoss tutorial)

Hi @sayakpaul! Apologies for the delayed response... So are you looking for the exact same cluster diagram or something similar (it's not a deterministic training so it won't always be the same)?

I've ran the tutorial and uploaded my embeddings here: https://drive.google.com/open?id=1ruK9P7LQHVMwvKlZYGi2NAawSxs99Z8g

I ran UMAP with default settings (15 neighbors and 2D) to get a pretty good clustering:

image

image

Closing but feel free to continue discussion here if there are any further questions or ways you think we can improve the tutorial.

sayakpaul

comment created time in 7 days

create barnchseanpmorgan/community

branch : gelu-migration

created branch time in 7 days

PR opened tensorflow/addons

Publish dev container

Closes #1414

Discussion Points:

  1. This adds a bit of computation time to the action on commit push. I don't think this is a big deal as this doesn't slow down feedback to contributors
  2. There is no addons repository included in the docker image. I believe this is the correct pattern since we want users to load their own repository within the container
  3. Because of 2 there is no pre-compiled bazel cache. We could add it but then the calls to bazel in the repository would need to be pointed to that cache. Given that our build only takes a couple of minutes I'd prefer to do this in another PR if we want to.

I tested the image in VScode using a dev-container and it worked as expected.

+71 -26

0 comment

7 changed files

pr created time in 7 days

push eventseanpmorgan/addons

Sean Morgan

commit sha f77e5d85a2f102ebdde759737266e3d76a1bf8cf

* Move to workflows since we need to run a command to clear space

view details

push time in 7 days

create barnchseanpmorgan/addons

branch : dev-container

created branch time in 7 days

push eventseanpmorgan/community

tongxuan.ltx

commit sha 9cdd68b1eb424a63972ca2f73f756882125727d5

RFC: FuseRecv

view details

Jeroen Bédorf

commit sha 5f54a3046413ef6027d3f3bdb438d232d18fe1f9

Fix irregular writing and expression issues.

view details

tongxuan.ltx

commit sha 57ee7b2ce16dcb9f0339c722f6c051ad6f6dd1d8

update reviewers and submit time

view details

tongxuan.ltx

commit sha b005a469afc335b839961557d7261dbf583a1998

update sponsor

view details

tongxuan.ltx

commit sha af8ab47f1962a03a77207c5a9630adb2e3c269e3

fix typo

view details

Thea Lamkin

commit sha 07661959690b4299686a2000f51a54442f80f4d2

Update CHARTER.md Added membership details, fixed typos

view details

ematejska

commit sha 912657cde917cca7216ed317ee82a94d8ee5dd26

Update to set the status to proposed.

view details

Thea Lamkin

commit sha 007219c18d55d247c4ec74e0e58441759ac829d8

Merge branch 'master' into theadactyl-patch-2

view details

Jiayi Zhao

commit sha 44159ab0ad4fbeca0e6cfa3b1a6e2340af79a38b

Add files via upload

view details

Jiayi Zhao

commit sha 3c390179e675eb2315b115bc9633c47ec949ea3a

Merge pull request #1 from 1025KB/1025KB-patch-2-1 1025 kb patch 2 1

view details

Jiayi Zhao

commit sha 877a03ab6472c5a2fcfc8437e26baa483fa4adbf

Delete cloud.png

view details

Jiayi Zhao

commit sha bd376ad8941a77a88fe9c3a2a8756707f050ea49

Delete parallel_tuning.png

view details

Jiayi Zhao

commit sha 352cce66d876bbc0c3d8c727db382ca03e7d342c

Delete workflow.png

view details

Jiayi Zhao

commit sha c1da0349088363383508ab53fd9e4b808ee1249c

Create 123

view details

Jiayi Zhao

commit sha 9cc6552fd03dc19e279386c082d3564c236791c8

Add files via upload

view details

Jiayi Zhao

commit sha 55a50fef7bc95651cf25b980ffb92d6df8bbf8d1

Delete 123

view details

Jiayi Zhao

commit sha 16fbadfa905ec0b56de8087969e50b3074f9dbac

Update 20200420-tfx-tuner-component.md

view details

Thea Lamkin

commit sha 7a2a5d58631aa147a6f9f14329b20dacce012412

Merge pull request #232 from tensorflow/theadactyl-patch-2 Update CHARTER.md

view details

tongxuan.ltx

commit sha f4b925cc72461f69adc16df9421b3fd5a4ad9155

Update user API and op definition based on comments.

view details

ematejska

commit sha d1c022d980a19b1e6e1f858adcbcfe7d9f1c4a8e

Merge pull request #236 from 1025KB/master RFC: TFX Tuner Component

view details

push time in 7 days

push eventseanpmorgan/addons

Sean Morgan

commit sha dfb69508b31f80eea4d966e1a05741b24841e3cd

* Move docker image to new namespace (#1881)

view details

Tzu-Wei Sung

commit sha c8ecee3191ccd95e52e353018d90d8b44ab995cf

Enable some compiler optimalization tricks for correlation cost GPU ops (#1855) * Enable some compiler optimization tricks * Add header * Fix missed input

view details

push time in 7 days

push eventtensorflow/addons

Tzu-Wei Sung

commit sha c8ecee3191ccd95e52e353018d90d8b44ab995cf

Enable some compiler optimalization tricks for correlation cost GPU ops (#1855) * Enable some compiler optimization tricks * Add header * Fix missed input

view details

push time in 7 days

PR merged tensorflow/addons

Enable some compiler optimalization tricks for correlation cost GPU ops cla: yes custom-ops
  • use ldg to read data
  • add __restrict__ qualifier to pointer
+40 -51

0 comment

1 changed file

WindQAQ

pr closed time in 7 days

Pull request review commenttensorflow/addons

Update and fix tutorials

         }       },       "source": [-        "try:\n",-        "    %tensorflow_version 2.x\n",-        "except:\n",-        "    pass\n",-        "\n",

Remove all try TF2.2 since it's default now

seanpmorgan

comment created time in 7 days

delete branch seanpmorgan/addons

delete branch : new-docker-namespace

delete time in 7 days

push eventtensorflow/addons

Sean Morgan

commit sha dfb69508b31f80eea4d966e1a05741b24841e3cd

* Move docker image to new namespace (#1881)

view details

push time in 7 days

more