I am Soamsree Majumder ,3rd year undergrad at IEM KOLKATA. I am looking forward to participating in GSOC 22 under this organisation. I am interested in ideas
“Create a model class for easier deployment PyMC models”
can you quantify how much experience you have working with PyMC so far?
@michaelosthege Right now I don’t have any pull request merged in pymc ,but I am working on few issues .Also I am well versed in python and Machine learning and had used pymc library in my project once or twice
Sir right now I am trying to go through the codebase and the wiki page of pymc so as to gain more insight about the same
Sounds good. I’m not quite up to date regarding the “Create a model class for easier deployment PyMC models” project.
Just want to give a little warning about it: This idea has been floating around for ~5 years now and the reason is mostly because it’s difficult.
While some models can be dumped with
cloudpickle, this is probably not what we’re looking for here.
My understanding of the project is that we’d like something similar to how, for example Keras stores deep learning models in HDF5. But for PyMC models which are Aesara graphs plus additional metadata.
So the challenge boils down to “How can an (arbitrary) Aesara graph be saved/loaded as something like HDF5/JSON etc.?”.