garrettwrong
commented
1 month ago Hi, thanks for your interest in ASPIRE. I'm not really able to follow your script or make out exactly what is happening from the log screenshot.
I would note that you are not using the most version of ASPIRE (12.2). A newer version is required for CUDA 12. I'm not sure what version you are attempting to run with here, or if you are reasonably installed... The current release and/develop versions are going to have the best support.
The closest script I have available to run is our cov3d demo, which looks to be where your script started from. I confirmed the following works on a machine with CUDA 12.3, and is automatically using the GPU for any available components with no user configuration required.
git clone git@github.com:ComputationalCryoEM/ASPIRE-Python ASPIRE-Python.main
cd ASPIRE-Python.main
conda create -n 1115 python=3.8 -y
pip install -e ".[dev,gpu-12x]"
python gallery/tutorials/tutorials/cov3d_simulation.pyI can generate errors if I manually override the config as you have. In that case the override is forcing portions of the code to run via cupy that have not actually been extended yet. As I mention above, you should get available GPU extensions automatically without changing the configuration. You may confirm this by noting the use of cufinufft in the logs, and also by invokingnvidia-smi from another shell session while the script it running to see the python process using the GPU.
Let me know if that works for you, thanks.
ThinkJanice
Describe the bug: Dear ASPIRE-Python Development Team, I am specifically working with the cov3d module from the ASPIRE package. However, after modifying the config.yaml file as shown below,
I encountered the following error.
Despite extensive research and troubleshooting attempts online, I have not found a viable solution to this issue. I apologize for the direct approach, but I would greatly appreciate any guidance or insights you could provide on this matter. Your support would be incredibly helpful to my research.
To Reproduce: import logging import numpy as np from scipy.cluster.vq import kmeans2 import starfile import pandas as pd import os
from aspire.basis import FFBBasis3D from aspire.covariance import CovarianceEstimator from aspire.noise import AnisotropicNoiseEstimator, WhiteNoiseEstimator from aspire.reconstruction import MeanEstimator from aspire.source.relion import RelionSource from aspire.utils import eigs from aspire.source.simulation import Simulation from aspire.denoising import src_wiener_coords from aspire.volume import Volume from aspire.utils.random import Random
from aspire.utils import RelionStarFile from aspire.operators import CTFFilter
logger = logging.getLogger(name)
Set input path and files and initialize other parameters
DATA_FOLDER = "xx" STARFILE = "xx" PIXEL_SIZE = 0.85 MAX_ROWS = None MAX_RESOLUTION = 16 CG_TOL = 1e-5 num_vols = 2 batchsize = 500
Set number of eigen-vectors to keep
NUM_EIGS = 32
Create a source object for experimental 2D images with estimated rotation angles, relionsource only cpu
print(f"Read in images from {STARFILE} and preprocess the images.") source = RelionSource( STARFILE, data_folder=DATA_FOLDER, pixel_size=PIXEL_SIZE, max_rows=MAX_ROWS )
source.images[:].asnumpy() = source.images[:].asnumpy().astype(np.float32)
print("The dtype of source is", source.dtype)
Downsample the images
print(f"Set the resolution to {MAX_RESOLUTION} X {MAX_RESOLUTION}") if MAX_RESOLUTION < source.L: source = source.downsample(MAX_RESOLUTION)
Specify the fast FB basis method for expanding the 2D images
basis = FFBBasis3D((MAX_RESOLUTION, MAX_RESOLUTION, MAX_RESOLUTION), dtype=source.dtype) print("The dtype of basis is", basis.dtype)
Estimate the noise of images
print("Estimate the noise of images using AnisotropicNoiseEstimator") noise_estimator = AnisotropicNoiseEstimator(source, batchSize=batchsize) #check also
Whiten the noise of images
print("Whiten the noise of images from the noise estimator") source = source.whiten(noise_estimator)
for l in range(0, batch_num):
source = np.load('{}{:04}_rwts_mat_l.npy'.format(ffbbasis_out,l), allow_pickle=True).astype(dtype)
Estimate the noise variance. This is needed for the covariance estimation step below.
noise_variance = noise_estimator.estimate() print(f"Noise Variance = {noise_variance}")
mean_estimator = MeanEstimator(source, basis = basis, batch_size=batchsize) # adjust the batch_size mean_est = mean_estimator.estimate()
Passing in a mean_kernel argument to the following constructor speeds up some calculations
covar_estimator = CovarianceEstimator(source, basis = basis, mean_kernel=mean_estimator.kernel) covar_est = covar_estimator.estimate(mean_est, noise_variance, tol=CG_TOL)
Extract the top eigenvectors and eigenvalues of the covariance estimate.
eigs_est, lambdas_est = eigs(covar_est, NUM_EIGS) for i in range(NUM_EIGS): print(f"Top {i}th eigen value: {lambdas_est[i, i]}")
Expected behavior: I hope to finish 3d mean and covariance estimation
Environment (please complete the following information):
develop]Additional context: Add any other context about the problem here.