Open kingfengji opened 7 years ago
@kingfengji
addig a mode.fit() and model.predict() is definitely a good thing to do if you want people to use it directly. Make sure you give guidelines or even create functions to save
and load
trained models!!!
I would recommend you to consider writing a python3 version as more and more people made the jump from 2.7 to 3.
I think these would be the more important improvement you can bring to the code with of course using users feedback to improve the code.
I will let you know if more important ideas come to my mind! Btw, feel free to look at my code to see how I created fit and predict functions (it's MIT license so free to use)
Best regards
Hi @pylablanche , Thank you very much for your suggestions, I'm new to this open-source thing and I appriciate your kindness. More readme and python3 then.
I've done some more improvement for speed via cython and numba, I'm happy to share some new results in the near future. Memeory issue is still a big thing, plan to tackle it next.
@kingfengji You're welcome! I'll happily give you more suggestions on the coming versions.
About the memory issue it is only a problem of slicing the pictures without copying it in the memory. Basically every slice would need to be a view and not a copy of the original picture. I've tried a couple of things but nothing really worked and I came to the conclusion that you need to control the way the picture is first stored in the memory. I'll give you more details if you are interested.
Best wishes
yep, python 3 support would be nice to have.
@pylablanche thank you very much for your tips. I was kind of really busy on some new research and sorry for late reply. Any suggestions / details are welcomed any time. thanks
@unnir I agree, python 3 should be included. and it will be..I just finished the sklearn-like api, will be out soon. thanks
Are you still working on Python 3 support? :)
While trying to install the requirements I encountered the problem that there's no tensorflow support below Python 3.5 for Windows machines.
Hi @pylablanche , I've just updated the package, feel free to use it and comment
Hi @kingfengji ,
That's great news!!! I'll definitely have a look and give you feedback if necessary.
For the next version, I plan to write some more wrappers such as model.fit() / model.predict() for a more user-friendly API if your goal here is to train the model out-of-box. (again, currently you actually can extract predictions)
Other suggestions is welcomed. Thanks.