Create AGENTS.md#2814
Conversation
|
|
||
| ## Docstrings | ||
| All functions/classes require a proper docstring. | ||
| Typing is a bonus but not mandatory. |
There was a problem hiding this comment.
How about this? In addition, this might be a good place to define a single docstring convention for pyFAI.
Would it make sense to adopt the NumPy style?
## Docstrings
All functions/classes require a proper docstring.
The docstring must describe the purpose, inputs, and outputs of the function in human-readable form.
Use NumPy-style docstrings if possible.
Typing hints are a bonus but not mandatory.
|
|
||
| ## Role | ||
| You are a senior developer. | ||
| Any modification you are requesting should be adequately tested and validated. |
There was a problem hiding this comment.
Suggestion
You are a senior scientific software developer.
Any modification you propose must be adequately tested and validated using pyFAI's testing framework.
Implementation details should go in comments, not in docstrings.
gudlot
left a comment
gudlot
left a comment
There was a problem hiding this comment.
A chance for unified docstrings?
Well, it is pretty low in the wish-list: most docstrings are in epydoc and got migrated gradually to sphinx. |
|
I still wonder if having such a file would:
For now it is far from being enough ! |
|
There is a comment on the usage of AI agents in #2824 (which was apparently done with them). This reactivates the interest for the topic. |
The comment says "This could enable Natural Language Processing (NLP) of Synchrotron/X-ray diffraction data, making the ecosystem highly accessible for users with less background in diffraction physics." To the best of my knowledge, NLP deals with semantics rather than diffraction data itself. The actual application here is instructing an agent to perform pyFAI tasks. For this to work correctly, a model would need to be trained to understand pyFAI instructions through correct prompt engineering and a high-quality model. I assume the existing pyFAI tests could be a valuable resource to help guide and validate such a model. A RAG, maybe knowledge based graph, would be the direction to go. |
2 participants
No description provided.