Will there be any assistant/office hours for Proj2 if we chose a ML4Science topic? Since these are "highly specific", do we get a designated assistant to help us work until the end?
For info, the lab we chose want us to read a paper, repurpose/change a given network and apply it to their own data, but thing is, we can't even start to get the demo.py working from the repo of the original paper, as the version may not be up to date, there are errors within the code, the pre-trained models are not available for download, etc.
you should use the thursday slots with the assistants for this. and the forum here can help too.
if you share more info about what kind of code/project it is here or on discord then we can also see which assistant is most suitable
Basically, we have received a paper about using a CNN for geometric matching, and got its associated GitHub repo (https://github.com/ignacio-rocco/cnngeometric_pytorch), and after reading the paper, personally I've spent many hours today trying to get into the code and trying to just get the demo.py to run.
This was because many class declaration/initialization used old, deprecated arguments, the pre-trained model weights weren't provided and I've had to check old commits to find the now deleted links that contained the pre-trained weights, figure out what the author was trying to do in his code. Now, having figured out some parts, I managed to get their demo_notebook.ipynb to run having modified many things such as filepaths, model constructor arguments, etc. and now wonder how it should be used on data other than the ones present in their notebook. I know I should refer to the readme, etc. but I have little faith in it, especially given how nothing worked as intended, and demo.py still doesn't run and I'm trying to debug every single line of code.
EDIT : obviously we will wait until Thursday to have any question, but just wondering if any assistant specifically has more experience with this topic.
ML4Science : Assistants/office hours
Hi,
Will there be any assistant/office hours for Proj2 if we chose a ML4Science topic? Since these are "highly specific", do we get a designated assistant to help us work until the end?
For info, the lab we chose want us to read a paper, repurpose/change a given network and apply it to their own data, but thing is, we can't even start to get the demo.py working from the repo of the original paper, as the version may not be up to date, there are errors within the code, the pre-trained models are not available for download, etc.
you should use the thursday slots with the assistants for this. and the forum here can help too.
if you share more info about what kind of code/project it is here or on discord then we can also see which assistant is most suitable
Here are some details
Basically, we have received a paper about using a CNN for geometric matching, and got its associated GitHub repo (https://github.com/ignacio-rocco/cnngeometric_pytorch), and after reading the paper, personally I've spent many hours today trying to get into the code and trying to just get the demo.py to run.
This was because many class declaration/initialization used old, deprecated arguments, the pre-trained model weights weren't provided and I've had to check old commits to find the now deleted links that contained the pre-trained weights, figure out what the author was trying to do in his code. Now, having figured out some parts, I managed to get their demo_notebook.ipynb to run having modified many things such as filepaths, model constructor arguments, etc. and now wonder how it should be used on data other than the ones present in their notebook. I know I should refer to the readme, etc. but I have little faith in it, especially given how nothing worked as intended, and demo.py still doesn't run and I'm trying to debug every single line of code.
EDIT : obviously we will wait until Thursday to have any question, but just wondering if any assistant specifically has more experience with this topic.
Add comment