just want to introduce a tool of Pytorch which helps me with the kaggle TGS challenge. It’s called Minetorch and was built by me during this competition, my friends and I use this the whole time with the TGS challenge and it does bring some conveniences for us.
Here’s the link https://github.com/louis-she/minetorch , it’s already on pypi so one can install it just with
pip install minetorch . Things out of box are:
- Tensorboard supported
- Auto resume
- Auto best model saving
- Many hook points for customization
There is also an mnist example of how to use this tool https://github.com/louis-she/minetorch/blob/master/minetorch/examples/mnist.py , it’s basiclly migrated from the official pytorch mnist example.
Sounds interesting? How does your tool compare to Ignite? Do you see any advantages / disadvantages using one or the other?
Haven’t known this until now, many features are same except the tensorboard and namespace.
I write this for my own need, not sure if it’s fit for everyone, just share it : )
Sure, thanks a lot for sharing it! Maybe some features would be nice to have in Ignite as well.
@ptrblck thanks for mentioning Ignite !
@chenglu sure that your project and Ignite have a lot of features in common. We are also working on more tight integration of Tensorboard into Ignite. If you have opportunity, give a try to Ignite and feel free to propose new features
Anyway nice analogy in your project presentation:
… So is data-mining, A special torch named pytorch can help us get the dimonds in data…
That’s great that PyTorch has a official tool like this, I will dig more about the Ignite and see if I can contribute. : )