You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
If we want to make this a broad NMR kit, we might get into issues with dependencies and their versions. Right now, I try to add a prediction model which was done previously. If I try to run it with a set of current software versions, I get an error, which seems to be due to a change in RDkit:
conformer = mol.GetConformer()[conf_i]
needs to become
conformer = mol.GetConformer(conf_i.item())
conf_i is a numpy type, the item() converts to int.
This workedin rdkit 2020.09.1.0
If I fix this, I get another error - the error comes from the code I want to import, and it is a wrong number of elements in a list. Whilst this is not directly due to versions, it does work in a different combination of library versions.
I think we will get into this issue more frequently - if we want to use existing code. Ideally, we would isolate the components. I am not an expert on docker, but can't docker do that (if we set it up)?
The text was updated successfully, but these errors were encountered:
Hello @stefhk3 thanks for pointing this out. I did think about this situation while designing this and our current solution is meant to avoid situations like this in its first place….
let’s say we resolve this issue in the current version and distribute the docker image you can always guarantee that it works on certain docker image release.
Here are some possible solutions
In a situation like this where we are packaging several components neither docker nor venv will be able to solve the problem because different components require different versions. It needs to be packaged as standalone docker images and call them explicitly from FASTAPI container and internally track the progress on another container to fetch and return the result of the run.
Other work around would be to install these packages manually and let the dependencies call them via aliases which is not advisable because we need to draw a limit otherwise this becomes non maintainable.
For now I still think we need to update the code to work with latest RDKit instead of the workarounds or complicating dependencies on multiple containers.
Do you already have a docker image with your model and RDKit running?. If so share the details here,
we can then run docker exec and run this in a separate docker image and be able to isolate everything.
If we want to make this a broad NMR kit, we might get into issues with dependencies and their versions. Right now, I try to add a prediction model which was done previously. If I try to run it with a set of current software versions, I get an error, which seems to be due to a change in RDkit:
conformer = mol.GetConformer()[conf_i]
needs to become
conformer = mol.GetConformer(conf_i.item())
conf_i is a numpy type, the item() converts to int.
This workedin rdkit 2020.09.1.0
If I fix this, I get another error - the error comes from the code I want to import, and it is a wrong number of elements in a list. Whilst this is not directly due to versions, it does work in a different combination of library versions.
I think we will get into this issue more frequently - if we want to use existing code. Ideally, we would isolate the components. I am not an expert on docker, but can't docker do that (if we set it up)?
The text was updated successfully, but these errors were encountered: