Efficientnet in Dataiku
Hello, I am currently working with Computer Vision problems and I need to implement a solution with efficientnet architecture.
I can see that with the plugins it can not be installed, also I tried to implement it with code but as Dataiku needs a version of TensorFlow below 2.0 and efficientnet is just in versions up to 2.0 I had no luck.
Has somebody implemented efficientnet in Dataiku? Or somebody knows whether this is possible, and if the answer is yes, can you tell me how could it be?
Operating system used: Linux Debian
Answers
-
Alexandru Dataiker, Dataiku DSS Core Designer, Dataiku DSS ML Practitioner, Dataiku DSS Adv Designer, Registered Posts: 1,212 Dataiker
Hi @luis_gmc
,You should be able to install efficientnet and tensorflow. I had not issues on Python 3.6 managed code env on DSS 9/10. We did see in the past that you may need to pin tensorflow ==2.1.0 in some cases.
For request packages I have :
Can you share the exact error you are seeing when trying to install efficientnet?
Thanks
-
luis_gmc Partner, Dataiku DSS Core Designer, Dataiku DSS ML Practitioner, Dataiku DSS Adv Designer, Registered Posts: 2 Partner
Hello, I've tried that and the error that I received is the following:
I think is due to the combination between keras with keras2
Thanks Alex
-
Alexandru Dataiker, Dataiku DSS Core Designer, Dataiku DSS ML Practitioner, Dataiku DSS Adv Designer, Registered Posts: 1,212 Dataiker
In terms of the error reported I see a similar report here :
Failed to train : <class 'ValueError'> : Input 0 is incompatible with layer flatten_1:expected min_ndim=3, found ndim=2
Can you check if that solution perhaps works for you?
Thanks,