Closed jS5t3r closed 2 years ago
It looks like an error produced by the PyNNDescent library that is used for approximate nearest neighbor search. The error messages are a little weird. Can you tell me which version of PyNNDescent and Numba packages are you using? I can try to reproduce the error.
pynndescent 0.3.3 py_0
numba 0.46.0 py37h962f231_0
all: https://gist.github.com/jS5t3r/63f418656f82ed1e5a6179c8b6b0883b
conda env export > environment.yml
: https://gist.github.com/jS5t3r/8ff34c52eea6753468035b0a8c025967 or
conda list -e > environment.txt
: https://gist.github.com/jS5t3r/b92c0e9998483151045eeed83929f2a9
you can use: conda create -n detection --file environment.yaml
can you just tell me the exact Python Version 3.x? This information is missing in your README.
@jS5t3r @jayaram-r I think it is Python 3.7.5. Please refer to this file
Thanks @machengcheng2016 I hope that resolved the error @jS5t3r
For future reference, using PyTorch 1.3.1 and other library versions given in the Readme file seems to solve the error. Works on a GTX 1080 TI, but not a TITAN V GPU.
To reduce the memory usage while extracting the layer embeddings, change the value of num_samples
on this line:
https://github.com/jayaram-r/adversarial-detection/blob/94fd0881a3eef179e66301629c9a5e348ce46bd1/expts/helpers/utils.py#L67
How I would set the params for FGSM:
python generate_samples.py -m cifar10 --aa FGSM --gpu 0
and then for layers.py