"ImportError: No module named neon.backends"

[rsbhat78]

(.venv2) bharagha@ubuntu-server:~/workload$ python deepspeech/speech/train.py --mainfest train:librispeech/train-clean-100/train-manifest.csv --manifest val:librispeech/train-clean-100/val-manifest.csv -e 4 -z 3 -s deepspeech/model_output.pkl -b mkl Traceback (most recent call last): File "deepspeech/speech/train.py", line 24, in from neon.backends import gen_backend ImportError: No module named neon.backends

I am trying to train using the librispeech samples. Ran into the above issue. Neon is installed as per the instructions on webpage. Kindly advise.

[rsbhat78]

I have been able to resolve the above issue. However, running into another ImportError issue with respect to the aeon.dataloader. This is happening though Aeon is installed fine. As a proof, I am able to run mnist example without any issue. Any hints will be very useful.

[tyler-nervana]

Could you post the stacktrace on the aeon error? Aeon is mostly used for large datasets like librispeech and imagenet, so the mnist example doesn't use it.

[rsbhat78]

(.venv2) bharagha@ubuntu-server:~/common/neon$ python ../deepspeech/speech/train.py --mainfest train:../librispeech/train-clean-100/train-manifest.csv --manifest val:../librispeech/train-clean-100/val-manifest.csv -e 4 -z 3 -s ../deepspeech/model_output.pkl -b mkl Traceback (most recent call last): File "../deepspeech/speech/train.py", line 23, in from aeon.dataloader import DataLoader ImportError: No module named aeon.dataloader

Attached the

[tyler-nervana]

Try re-installing aeon from within your neon virtual environment (see here). It should just require these three steps:

git clone https://github.com/NervanaSystems/aeon.git
cd aeon
python setup.py install

[rsbhat78]

Thanks. It helped.

[rsbhat78]

One more question. I am trying to execute the following command: _python ../../deepspeech/speech/train.py --manifest train:/home/bharagha/common/librispeech/train-clean-100/train-manifest.csv --manifest val:/home/bharagha/common/librispeech/train-clean-100/val-manifest.csv -e 2 -z 3 -s ../deepspeech/modeloutput.pkl -b mkl

I get a segmentation fault with the following message:

2017-08-07 19:04:35,828 - neon.util.argparser - WARNING - No schedule given for model serialization, using default 1 2017-08-07 19:04:35,874 - neon.backends - WARNING - deterministic_update and deterministic args are deprecated in favor of specifying random seed [src/conv.c:151] err (-127) [src/conv.c:152] err (-127) [src/conv.c:153] err (-127) [src/conv.c:172] err (-1) [src/conv.c:173] err (-1) [src/conv.c:174] err (-1) [src/conv.c:175] err (-1) [src/conv.c:176] err (-1) [src/conv.c:177] err (-1) [src/conv.c:178] err (-1) [src/conv.c:179] err (-1) [src/conv.c:180] err (-1) [src/conv.c:204] err (-127) [src/conv.c:220] err (-1) Segmentation fault (core dumped)

Any pointers?

[rsbhat78]

Guess this happened because there is no support for MKL. Setting backend to CPU works. Thanks.