Cannot run demo in pycharm(2019.1) in windows 10. about demo-routenet HOT 9 CLOSED

Hello,
We tested the code in Windows 10 with PyCharm and it worked. Probably, you have a problem with the library dependencies or the CUDA distribution. Please make sure you are using the same versions of the libraries listed in the "requirements.txt" file.

Regards,
José

from demo-routenet.

Thanks for replying in a short time, i checked the installed libraries for "requirements.txt" file once again, it matches requirements.txt. For my test in windows, i installed CUDA(10.0 ver.) and CUDNN(7.6.4 ver.), so i wonder if there is something different from your CUDA&CUDNN version. Thanks, Wayne

from demo-routenet.

Hello Wayne,
Please check you are using TensorFlow v1.14. In principle, with this version you should use CUDA v10.0 and CUDNN v7.4. Probably, you are not using a valid CUDNN version.

Also, remember adding the environment variables in Windows:
# SET PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0\bin;%PATH%
# SET PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0\extras\CUPTI\libx64;%PATH%
# SET PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0\include;%PATH%
# SET PATH=C:\tools\cuda\bin;%PATH%

Regards,
José

from demo-routenet.

Hi José, i re-installed CUDNN7.4.2 for testing, but still the demo cannot run. i debuged routenet_with_link_cap.py in pycharm, it showed like this:

Instructions for updating:
Use standard file utilities to get mtimes.
I0326 11:44:22.388325 4452 session_manager.py:500] Running local_init_op.
I0326 11:44:22.478085 4452 session_manager.py:502] Done running local_init_op.
I0326 11:44:28.776959 4452 basic_session_run_hooks.py:606] Saving checkpoints for 0 into ./CheckPoints/nsfnetbw\model.ckpt.
W0326 11:45:50.676099 4452 ops.py:465] Tensor._shape is private, use Tensor.shape instead. Tensor._shape will eventually be removed.
W0326 11:45:50.701011 4452 ops.py:465] Tensor._shape is private, use Tensor.shape instead. Tensor._shape will eventually be removed.

Process finished with exit code -1073741819 (0xC0000005)

from demo-routenet.

In pycharm terminal, i put this:
python routenet_with_link_cap.py train --hparams="l2=0.1,dropout_rate=0.5,link_state_dim=32,path_state_dim=32,readout_units=256,learning_rate=0.001,T=8" --train d:/omnetpp-6.0pre7/samples/routenet/nsfnetbw/tfrecords/train/.tfrecords --train_steps 50000 --eval_ d:/omnetpp-6.0pre7/samples/routenet/nsfnetbw/tfrecords/evaluate/.tfrecords --model_dir ./CheckPoints/nsfnetbw

Is this command right? i wonder the running failure is because of the wrong "--train " parameter and "--eval_" parameter in windows10. That means the demo cannot get "*.tfrecords" from the right file path.

from demo-routenet.

Hi Watson,
Yes, the command looks right. I assume that the paths to "train" and "eval_" include a wildcard at the end ("...*.tfrecords") but it was removed due to the Markdown language.

I would try to execute it from the Windows terminal just to discard it is a problem with PyCharm.

Regards,
José

from demo-routenet.

Hi José, i did some tests for the wildcard("....tfrecords") , it turned out to be the problem with Markdown. More over, i changed the "train" and "eval_" as input parameters($4 and $5) like this:
sh run_routenet.sh train nsfnetbw 50000 d:/tfrecords/train/.* d:/tfrecords/evaluate/.

Also, some changes in run_routenet.sh:
if [[ "$1" = "train" ]]; then

python routenet_with_link_cap.py train --hparams="l2=0.1,dropout_rate=0.5,link_state_dim=32,path_state_dim=32,readout_units=256,learning_rate=0.001,T=8" --train_steps $3 --train $4  --eval_ $5 --model_dir ./CheckPoints/$2

fi

However, when i checked the file "graph.pbtxt", it could only find one file(maybe the first one) in "nsfnetbw\tfrecords\train", here it is:
attr {
key: "value"
value {
tensor {
dtype: DT_STRING
tensor_shape {
dim {
size: 1
}
}
string_val: "d:/tfrecords/train/results_nsfnetbw_10-12_Routing-SP-k-0_0_124.tfrecords"
}
}
}
}
Best regards,
Wayne

from demo-routenet.

Hi Wayne,
Please note that this demo is prepared to use the datasets we provide here ('datasets_v0').
Looking at the names of your tfrecords I see you are using 'datasets_v1', which have a different structure. In this case, you would need to modify the code to read correctly the tfrecords.

Regards,
José

from demo-routenet.

Yes, i did use datasets_v0 for testing. For datasets_v0, i ran it well in Ubuntu, but for the reason of wildcard("/*.tfrecords"), the demo couldnot get the right path for xx.tfrecords files in pycharm terminal(windows cmd). For the testing above, i put "train" and "eval_" as input parameters($4 and $5) trying to solve the problem of wildcard, so i am puzzled why it found only one .tfrecords file in "graph.pbtxt".

from demo-routenet.