I am aware that this question can be answered in different ways based on the domain of research under consideration. So, you can confine the answer to the domain of machine learning.
Suppose I am working on a technique in a particular application using a model and get success. And there is a certain hope that the model can further be improved drastically. I don't want to put any of my papers on any pre-print.
In this context, I got a dilemma on the question "should I wait till complete experimentation for publishing or should I announce the success first and then improve?"
Pros and cons for waiting:
- I can write a full-fledged research paper.
- No need of doing an incremental paper.
- I cannot announce the results quickly.
- I cannot use my results for my academic requirements if someone publishes the same early.
- Others may treat it as a salami-slicing publication.
Note: Complete experimentation or full-fledged refers to developing a fine-tuned model. But the experimentation showing the domination of the (model based on) technique is already established.