Open harrytchild opened 1 month ago
I noticed that in the CPU run you used the full paths for the output dir (/localdata/harry/MAF024/unsterilised_coassembly_output/bins/M_U/graphmb_results) and in the GPU run you used relative paths: (gpu_test_graphmb_results) I would have to check what's going on with the "writebins" option to break there with relative path, but can you try again with full paths as you did in the CPU run, to make sure that's the issue? Thanks!
Hi Andre,
Thanks for getting back to me. I have tried using full paths and this still gave the same error.
Cheers,
Harry
Hi Andre,
I have managed to run GraphMB using CPUs but this took a long time on the large assemblies I have. I had issues running the same installation on the GPU, which I thought I had fixed by changing some of the dependancy versions (see below). However, I have now come across another issue using this installation on the same dataset. Here is the command I have used:
graphmb --assembly metaflye_assemblies/H_U --outdir gpu_test_graphmb_results_H --assembly_name edges_filt.fasta --markers assembly_contig_marker_gene_stats.tsv --graph_file assembly_graph.gfa --depth edge_depth.txt --writebins --cuda
And the error:
To Reproduce Steps to reproduce the behavior:
Environment used
Type of data Here are the steps to the input data
How GraphMB was installed mamba create -n graphmb -c conda-forge make cmake libgcc python=3.9 pip tensorflow=2.11.1 numpy=1.23.0 mamba activate graphmb pip install graphmb
Options used See command above
Expected behaviour Here is the log from running GraphMB on the same files with CPU: cpu_graphmb_log.txt
Output messages Here is the log file with the error: gpu_test_graphmb.txt
Cheers in advance for any help you can provide!
Harry