0 votes

I have some problems with using the Deep Learning Image plugin:

  1. The CPU-plugin installs, but does not 'publish' any recipes (e.g. the 'Recipe: Image classification').
  2. The GPU plugin installs with the recipes. However:
    1. Using the 'Image Classification'  recipe results in an (HTTP 500) error in the dialog screen (see below). Classifying the images seems to work though.

    2. Using the 'Retraining Image Classification Model' also results in an HTTP 500 error, but this time I cannot change the 'Image filename column' and the 'Label column' to the required settings. Retraining results in errors.

 

How do I install the CPU plugin correctly and what do I need to do to get the Retraining recipe working?

 

Thanks!

Peter.

asked by
Solved!
Thnx Nicolas for your instructions and your help.

Regards,
Peter

1 Answer

+1 vote

Hello Peter,

There was indeed a bug with the plugin due to the recent release of the package pip (version 9.0.2) which seems to be broken in the context of the plugin. We fixed the issued and released new versions of the plugin (both CPU and GPU) : 0.1.2

You can update your version directly from the Plugin Store inside DSS, in Administration > Plugins, and clicking on Update to 0.1.2

Be sure to install the external dependencies as proposed by DSS. They are mandatory for the plugin to work. If you already installed the ones for the previous version, click on change to update them. Then, you will see that the selected code env is not up to date. You need to create a new one.

To see the recipes, you can refresh or even force refresh your browser to be sure that they are available.

If you want to use the GPU version of the plugin, you need to have access to a NVIDIA graphics card, with CUDA 8.0 and cudNN 6.0 installed.

Don't hesitate if you any other questions,

Best,

Nicolas

 

answered by
Hi Nicolas,

Version 0.1.2 of the Deep Learning plugins is not (yet) available in the store...

Regards,
Peter
Hello Peter,

You can refresh the store list by clicking on the refresh button on the right of the ADVANCED button at the top left of the Store screen.

Then you should be able to see version 0.1.2.

Best,

Nicolas
Hello,
Upgrading to version 0.1.2 does not seem to solve the problem.
I did create a new (managed) code environment while updating.
However, the recipes are still not available in the CPU-version of the plugin.

Regards,
Peter.
Hello Peter,

it seems that your installation of the plugin is corrupted because I cannot reproduce this issue.

If you don't need the GPU version of the plugin, a solution to this problem is to delete the plugins (both the GPU and CPU versions) from DSS and to re-install only the CPU version.

Unfortunately, it is not possible to do that from the DSS UI. Do you have access to the instance where DSS is installed, and in particular to the DATADIR ( more information on this here: https://www.dataiku.com/learn/guide/getting-started/dss-concepts/the-dss-datadir.html ) ?

If yes, you can:
- Stop your DSS instance (by running the following command: DATA_DIR/bin/dss stop, make sure that there are no current jobs running at that moment)
- Go to DATADIR/plugins/installed and remove the 2 plugins folders (deeplearning-image-gpu and deeplearning-image-cpu)
 - Restart your DSS instance (by running the following command: DATA_DIR/bin/dss start)
- Reinstall the CPU version of the plugin from the Store

You should have only one version of the plugin installed on a particular DSS instance. If you have access to a GPU, you can install the GPU version, and then decide in each recipe whether you want to use the GPU or the CPU to run them.

Best,

Nicolas
919 questions
956 answers
953 comments
1,777 users