add GPU option for KNN classifier, update requirements.txt

[chaudatascience]

Some small updates on the evaluation part:

• Add enviroment.yml for setting up Conda Environment for the project
• Add GPU option for KNN classifier
• Minor format updates.

[Zitong-Chen-16]

Thank you for catching the typos and including the gpu options! It looks good to me. My only question is for the environment setting. Juan and I discussed earlier that we might want to use pip instead of conda which is why I included the requirement.txt file, but of course conda is also a viable option. Should we just use conda instead @jccaicedo ?

[chaudatascience]

Good point! Using pip may be a better choice here if someone only wants to check their model's performance (on their current env) and doesn't want to switch to a new env to just evaluate it. I'll revert back to the requirements.txt.

[chaudatascience]

I have reverted back to using pip and have tested the installation on my machine.

[jccaicedo]

Looks good to me!

[jccaicedo]

Thank you for the PR, @chaudatascience ! @Zitong-Chen-16 can you review and test on your env? After it works, feel free to merge!

[Zitong-Chen-16]

Hey @chaudatascience! I tried running the updated code but ran into an issue. The code ran successfully without the gpu though. But I ran the code with the gpu option (i.e. with --use_gpu), I received the following error. Were you able to run it on your machine with the gpu option?

Faiss assertion 'err == CUBLAS_STATUS_SUCCESS' failed in void faiss::gpu::runMatrixMult(faiss::gpu::Tensor<float, 2, true>&, bool, faiss::gpu::Tensor<T, 2, true>&, bool, faiss::gpu::Tensor<IndexType, 2, true>&, bool, float, float, cublasHandle_t, cudaStream_t) [with AT = float; BT = float; cublasHandle_t = cublasContext*; cudaStream_t = CUstream_st*] at /project/faiss/faiss/gpu/utils/MatrixMult-inl.cuh:265; details: cublas failed (13): (512, 1536) x (31060, 1536)' = (512, 31060) gemm params m 31060 n 512 k 1536 trA T trB N lda 1536 ldb 1536 ldc 31060

[chaudatascience]

Yes. I was able to run it with GPU.
I tested by creating a new conda environment and installing the packages listed in the requirements.txt as follows

conda create -n test_env python=3.9
conda activate test_env
cd MorphEm
pip install -r requirements.txt
python run_benchmark.py --use_gpu

Can you try again to see if the error still persists?

hmm, it seems that some people have encountered a similar problem, as reported link1 and link2.
The author recommends using Conda conda install -c pytorch faiss-gpu to install it. If the one above doesn't work, maybe using conda to install can help.

[Zitong-Chen-16]

That solved the problem! I did the exact same thing to create the environment and pip install via the file earlier and ran into the issue, but conda install faiss-gpu and pytorch seemed to fix it.

[Zitong-Chen-16]

Can I merge the pull request now then @chaudatascience ? Maybe we could update the readme to include the conda install faiss-gpu after

[chaudatascience]

Yeah, I think we can update it later.

[Zitong-Chen-16]

Sounds good! Thank you!!