kirthevasank / nasbot Goto Github PK

The code size is too large, not easy to read, how to use random search in the source code?Can you give an example? Thank you.

Dear @kirthevasank,

Thanks for your great implementation. I try to run the code with demos/demo_cnn.py. But I see X1, X2 is empty to compute the distance, leading to a bug in computation.
In file opt/nasbot.py
init_gp_fitter = NNGPFitter([], [], self.domain.get_type(), tp_comp=self.tp_comp,
list_of_dists=None, options=self.options,
reporter=self.reporter)

Could you help to solve this issue? Thanks so much.

hi, I meet the problem when i follow the step 'bash make_direct.sh' & ' python simple_direct_test.py '
I have installed both python2 and python3 on my computer, after the 'bash make_direct.sh',
'ImportError: No module named direct' is occured when I use 'python2 simple_direct_test.py',
but it's work right "python3 simple_direct_test.py", what should I try to make it work on python2? Thanks!

Thanks for releasing all of your code for this. Could you make a setup.py so that this is installable via pip? It seems pretty close.

FYI I'm looking to use the OTMANN distance in a separate project, but the local references are making it difficult for me to include just the distance without copying over all of the files and updating the paths.

SliceLocalization. P is unreadable and always reports errors：
ValueError: insecure string pickle.
It was already dumped. Forget to close when saving：https://stackoverflow.com/questions/1746825/valueerror-insecure-string-pickle
Can you give a new one ?

Hi!,@kirthevasank, I have forked the project and run it in the cifar10 dataset, all the architectures searched by the Nasbot have low accuracy compare to the SOTA architectures. I have changed some HyperParameters, but it's still not work, can you guess the reason for the "lower" accuracy ? Thanks!!

Hi!
Just as your paper, the Cifar-10's error rate curves could be plot as this.

However, When I tried to run your code, I got this curves.

I want to know if there is something wrong

Hi,

If we want to find architectures on big datasets such as ImageNet, how to change our code to support it?

Thanks