Icevision: Improper Packaging with fastai requirement

[oke-aditya]

Re opened from #214

@rashmimarganiatgithub
cc @lgvaz

Let's solve this here. So that other people too can benefit if they face this.

We are trying to package our code properly for deployment still. #217.

@lgvaz This improper packaging is a concern. #181 problem.

Really sorry @rashmimarganiatgithub this package is still preparing for its first release 0.1 and we yet need to fix such issues.

[rashmimarganiatgithub]

@oke-aditya I saw your previous comments here: "That means in Kaggle competitions dataset. Upload fastai v2, PyTorch lightning wheels, install them then try to install mantisshrimp."
Hope I have installed it properly. here is the link.
https://colab.research.google.com/drive/1btLZqnRsZZgd1roF1RJURAeqHY8k7ruO?usp=sharing

Datasets details:

https://www.kaggle.com/kaushal2896/pycocotools
https://www.kaggle.com/geethasaikrishna/fastai-installer
https://www.kaggle.com/validmodel/withwhl
https://www.kaggle.com/ar90ngas/dogs-vs-cats-efficientnet-requirements

[rashmimarganiatgithub]

Re opened from #214

@rashmimarganiatgithub
cc @lgvaz

Let's solve this here. So that other people too can benefit if they face this.

We are trying to package our code properly for deployment still. #217.

@lgvaz This improper packaging is a concern. #181 problem.

Really sorry @rashmimarganiatgithub this package is still preparing for its first release 0.1 and we yet need to fix such issues.

Can it be fixed within 1 or 2 days from now?. Only 9 days are remaining.

[oke-aditya]

We can fix it ASAP, can you share your kernel !!. @lgvaz and I can fix it up and hand over kernel to you.

[lgvaz]

@rashmimarganiatgithub As I said in #214 , we do need a reproducer for the error, or else it's going to be very hard for us to fix the issue

[rashmimarganiatgithub]

We can fix it ASAP, can you share your kernel !!. @lgvaz and I can fix it up and hand back kernel to you.

In order share the kernel you should be part of the team. Hence it is against the rules to share with other, so I have downloaded the code and shared you in google colab and will share the Kaggle dataset which I am using as the reference to fix it.

[rashmimarganiatgithub]

@oke-aditya I have updated the comment with datasets. Hope it helps you to reproduce the bug easily.

[oke-aditya]

@lgvaz we need to have a look.

[rashmimarganiatgithub]

@oke-aditya any improvement on this?

[oke-aditya]

@rashmimarganiatgithub
Yes, I am trying to create a proper wheel and packaging with #226 also #223 was made to fix it.
I have uploaded the wheel file of mantisshrimp on kaggle dataset for now. I will look into the installation as you have put. We will share a kernel/notebook that shows you how to install this package with internet off and wheels.
I understand that competition is close to end maybe a week left. But we are trying best to get there !!

[lgvaz]

@rashmimarganiatgithub

I updated the dependencies so you don't need to install pytorch-lightning

If I understood your problem correctly, you only need to do predictions offline, if that's the case, you don't need to install fastai, just fastcore will suffice.

In your notebooks you were doing:

from mantisshrimp.models.rcnn.faster_rcnn import *
from mantisshrimp.models.rcnn import *

That can cause problems with our soft-dependencies, the correct way of importing is:

from mantisshrimp.models.rcnn import faster_rcnn

Here is a minimal installation notebooks, I hosted the wheel used there here

Please tell us if everything is working now =)

[rashmimarganiatgithub]

@rashmimarganiatgithub

I updated the dependencies so you don't need to install pytorch-lightning

If I understood your problem correctly, you only need to do predictions offline, if that's the case, you don't need to install fastai, just fastcore will suffice.

In your notebooks you were doing:

from mantisshrimp.models.rcnn.faster_rcnn import *
from mantisshrimp.models.rcnn import *

That can cause problems with our soft-dependencies, the correct way of importing is:

from mantisshrimp.models.rcnn import faster_rcnn

Here is a minimal installation notebooks, I hosted the wheel used there here

Please tell us if everything is working now =)

@lgvaz @oke-aditya it is working... thanks much..