Models: [domain adaptation] no such package 'domain_adaptation/datasets'

[JJTasi]

System information

-What is the top-level directory of the model:
GPU-Server1:~/domain_adaptaion$ ls
datasets domain_separation __init__.py README.md WORKSPACE

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04):16.04
• TensorFlow installed from (source or binary):
• TensorFlow version (use command below):1.0
• Bazel version:0.4.5
• CUDA/cuDNN version:8.0
• GPU model and memory: GeForce GTX TITAN 6GB

I'm facing this issue-
When running
bazel build -c opt domain_separation/...

I get this error:
ERROR: /home/jaytest/domain_adaptaion/domain_separation/BUILD:103:1: no such package 'domain_adaptation/datasets': BUILD file not found on package path and referenced by '//domain_separation:models_test'. ERROR: Analysis of target '//domain_separation:models_test' failed; build aborted. INFO: Elapsed time: 0.983s

And I run again
bazel build -c opt domain_separation/...

I get another error:
ERROR: /home/jaytest/domain_adaptaion/domain_separation/BUILD:79:1: no such package 'domain_adaptation/datasets': BUILD file not found on package path and referenced by '//domain_separation:dsn_train'. ERROR: Analysis of target '//domain_separation:dsn_train' failed; build aborted. INFO: Elapsed time: 0.096s
I already checked these files are under the "domain_separation" folder

Can you please help me how to resolve it?

[girving]

Cc @bousmalis.

[bousmalis]

Hi, you will need to build from the parent directory (models) that includes the slim and domain_adaptation directories. From there you can follow the instructions: https://github.com/tensorflow/models/tree/master/domain_adaptation

Also note that both models and models/slim need to be added to your $PYTHONPATH. Please, let me know if there are any more issues.

[girving]

@bousmalis If the bazel command depends on the working directory in which it is executed, that's seems like a bug that should be fixed.

[JJTasi]

Thanks for your help.
I can run it without error now.
But, I have another problem.
I follow instruction in slim and get mnist_train.tfrecord file.
Then, I modify the downloading and converting code in slim and get mnist_m_train.tfrecord file.
And I apply DSN on these two dataset by following command

./bazel-bin/domain_adaptation/domain_separation/dsn_train --similarity_loss=dann_loss --basic_tower=dann_mnist --source_dataset=mnist --target_dataset=mnist_m --learning_rate=0.0117249 --alpha_weight=0.01 --beta_weight=0.05 --gamma_weight=0.251175 --weight_decay=1e-6 --max_number_of_steps=60000 --layers_to_regularize=fc3 --use_separation=True --master="" --dataset_dir=/home/jaytest/models/domain_adaptation/datasets -v --use_logging --train_log_dir=/home/jaytest/models/domain_adaptation/tmp/da

I can only get test accuracy about 72% in mnist_m.

Is there any hyperparameter that i didn't set up correctly?

[bkkm78]

@JJTasi I am having the same problem here. Test accuracy is around 0.72. Have you further investigated this issue?

[xiaochenwang94]

I run the command bazel build -c opt domain_adaptation/domain_separation/... at research directory, this directory is the parent directory of slim and domain_adaptation. But I still have the error no such package 'domain_adaptation/datasets'. I clone the master branch of this repo, how can I solve this problem?

[karmel]

@bousmalis can you comment?

[HutzelFutzel]

Same problem here. Is there a solution for it?

[475000327]

Hello,everyone,I also come across the same question.Running the command bazel build -c opt domain_adaptation/domain_separation/... at research directory, this directory is the parent directory of slim and domain_adaptation. But there still have the error no such package 'domain_adaptation/datasets'.
Can anyone please help me how to resolve it?

[AmirHussein96]

I reproduced a simplified version of DSN, you can check it here https://github.com/AmirHussein96/Simplified-DSN

[475000327]

OK～wonderful！Really thank you for your code and your reply～

------------------ Original ------------------
From: Amir Hussein <[email protected]>
Date: Thu,Oct 31,2019 4:39 AM
To: tensorflow/models <[email protected]>
Cc: 475000327 <[email protected]>, Comment <[email protected]>
Subject: Re: [tensorflow/models] [domain adaptation] no such package 'domain_adaptation/datasets' (#1396)

I reproduced a simplified version of DSN, you can check it here https://github.com/AmirHussein96/Simplified-DSN

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub, or unsubscribe.

[jaeyounkim]

@bousmalis @dmrd Can you add the link of AmirHussein96's repo to the README.md file so that TF community can find an unofficial version easily?

[jaeyounkim]

Added the link of AmirHussein96's repo to the README.md file.
https://github.com/tensorflow/models/blob/master/research/domain_adaptation/README.md#other-implementations

@AmirHussein96 Thank you for your contribution.