Welcome to TomoTwin Discussions! #15
Replies: 6 comments 13 replies
-
|
Hi Thorsten, Is it possible to get Tomotwin to work on Mac m1 ? Thanks |
Beta Was this translation helpful? Give feedback.
-
|
Mh, havent tried it yet. But pyTorch recently added M1 support. So it might be possible but is not a priority right now. |
Beta Was this translation helpful? Give feedback.
-
|
Hi Thorsten, Great software ! Working perfectly on a Linux environment. I wanted to know if it is possible to increase the number of CPUs when running the "tomotwin_locate.py findmax" command. On a more general comment, how can we know more about the parameters we feed to the software during the different steps?
Thanks in advance, |
Beta Was this translation helpful? Give feedback.
-
|
Well, I just had a look at the computation I started yesterday, and it seems that at some point tomotwin uses more than 4 CPUs (see first attached screenshot). But see how 3 jobs are much older than the other ones? Those are the ones I started yesterday, there is one missing though, because yesterday I could see 4 of them... And I guess that when at least one of the jobs finished, then it unleashed the whole potential of the computer. Though I do not know why there are so many instances of tomotwin running.... I count 27 of them, give or take. Is this a normal behaviour?
|
Beta Was this translation helpful? Give feedback.
-
|
Thanks for getting in touch!
Cool! We are always curious how TomoTwin does in the wild. If you want, please share some details how the results look like. If its confident, you can also drop us a PM.
You can list all options with: the option However, it's only parallelized over the references. So a single reference won't be processed faster by using more processes.
Does the output of the following command answer you question?
The findmax procedure is one way to detect the peaks in the 3D heatmap. Other procedures are also possible (like using thresholding and connected components analysis), therefore I capsuled it into a seperate command. So in the future there might a command like Does that answer your questions? :-) |
Beta Was this translation helpful? Give feedback.
-
|
Hi Thorsten,
Yeah, at some point I will definitely be able to share some results, I just want to play a bit with the software first.
I will try that as soon as I can (the server is currently down), thanks heaps!
Yep, this is what I figured, but since I had 12 references I knew I could use more CPU ;-)
It answers them perfectly! Thanks for that! |
Beta Was this translation helpful? Give feedback.
-
|
Hey! I'm really psyched to try tomotwin, but hit a little snag: trying to load the umap embedding in napari leads to the error below. Has anybody else seen this, too? O:-) Thanks, Philipp. AttributeError Traceback (most recent call last) File ~/miniconda3/envs/napari-tomotwin/lib/python3.10/site-packages/psygnal/_weak_callback.py:291, in _StrongFunction.cb(self=<psygnal._weak_callback._StrongFunction object>, args=()) File ~/miniconda3/envs/napari-tomotwin/lib/python3.10/site-packages/magicgui/widgets/_function_gui.py:218, in FunctionGui.init.._disable_button_and_call() File ~/miniconda3/envs/napari-tomotwin/lib/python3.10/site-packages/magicgui/widgets/_function_gui.py:331, in FunctionGui.call(self=<FunctionGui load_umap_magic(label_layer: napari...Tomo_009-dose_filt_embeddings.tumap'))>, update_widget=False, *args=(), **kwargs={}) File ~/miniconda3/envs/napari-tomotwin/lib/python3.10/site-packages/napari_tomotwin/load_umap.py:34, in load_umap_magic(label_layer=None, filename=PosixPath('Tomo_009-dose_filt_embeddings.tumap')) AttributeError: 'NoneType' object has no attribute 'opacity' The above exception was the direct cause of the following exception: EmitLoopError Traceback (most recent call last) File ~/miniconda3/envs/napari-tomotwin/lib/python3.10/site-packages/psygnal/_signal.py:935, in SignalInstance.emit(self=<SignalInstance 'changed' on PushButton(value=False, annotation=None, name='call_button')>, check_nargs=False, check_types=False, asynchronous=False, *args=(False,)) File ~/miniconda3/envs/napari-tomotwin/lib/python3.10/site-packages/psygnal/_signal.py:982, in SignalInstance._run_emit_loop(self=<SignalInstance 'changed' on PushButton(value=False, annotation=None, name='call_button')>, args=(False,)) EmitLoopError: calling <psygnal._weak_callback._StrongFunction object at 0x7fab70eecc10> with args=(False,) caused AttributeError: 'NoneType' object has no attribute 'opacity'. |
Beta Was this translation helpful? Give feedback.
-
|
Hi, sorry to hear that you have issues :-) To make sure that we are using the same napari version could you run: In case you don't have After you did that run it again. I case it crashes again send me the Best, |
Beta Was this translation helpful? Give feedback.
-
|
Thank you so much for the support! :-) I actually did the update just before. But let me do it again! The problem still persists. Here's the list of installed packages: packages in environment at /home/philipp/miniconda3/envs/napari-tomotwin:Name Version Build Channel_libgcc_mutex 0.1 conda_forge conda-forge |
Beta Was this translation helpful? Give feedback.
-
|
Ah! Found it ... so the mask is necessary? |
Beta Was this translation helpful? Give feedback.
-
|
But I also confirmed on another system that there's a problem with the PyQt bits label_layer.visible = 1 |
Beta Was this translation helpful? Give feedback.
-
|
sorry, I totally forgot you!! Let me try to reproduce it... |
Beta Was this translation helpful? Give feedback.
-
|
So here is what I did: Then: Plugins -> Cluster UMAP Embeddings -> Select File (umap) -> Load I dont see any problem here. And yes, you need to provide the label_mask! |
Beta Was this translation helpful? Give feedback.
-
|
Alright, with the next release there be pop-up message if parameters are not specified: |
Beta Was this translation helpful? Give feedback.
-
When does that happen? I need to reproduce it to fix it. |
Beta Was this translation helpful? Give feedback.
-
|
Lets we continue here: |
Beta Was this translation helpful? Give feedback.
-
|
Hi Thorsten, Thank you for making tomotwin! We really like it, and so far have used it succesfully on a small dataset for STA of a protein bound to liposome membranes (seems to do a good job, even though the model wasn't trained on this). My test dataset was 10 tomograms, which was manageable. We now have 180 tomograms to process - and embedding is a major bottleneck. We will try the masking outlined in the latest version. Thanks again, |
Beta Was this translation helpful? Give feedback.
-
|
Hi Iain, Thanks for sharing your experience with TomoTwin. We know that embedding is quite slow and a bottleneck. We are working hard on this issue. May I direct your attention to this PR: With this PR the embedding will be much faster. Additionally it much better scales with multiple GPUs (how many do you use?). The work is basically done, I'm just testing if everything works as it should because the internal libs have been updated. If you want, I can point you to the installation instructions for the beta and you give it a try? External testing would also help us :-) Regarding the step: Hope this helps you! Best, |
Beta Was this translation helpful? Give feedback.
-
|
Hi Thorsten, Thanks for the quick reply and all the helpful info.
Generally I have been using 4 GPUs, but can request a max of 8 from our cluster (current plan was to split tomograms into 2 sets running simultaneously with 4 GPUs for each job) - so this may make things very fast. I would be happy to do some testing!
Ah OK - the protein is large (~ 30 x 15 pix in 2D in the 10A bin tomo), so maybe some wiggle-room there. Thank you! Best wishes, |
Beta Was this translation helpful? Give feedback.
-
👋 Welcome!
We’re using Discussions as a place to discuss ideas or questions around TomoTwin. :-)
If you have certain problems with TomoTwin we can also start here and if needed convert it into an issue.
Lets get in touch :-)
Beta Was this translation helpful? Give feedback.
All reactions