Allow dynamic parameters as dvc pipeline dependence

[VGODIE]

I have faced an issue when I need to repro pipeline when new week starts - because new week brings more data to fit the model on. However, it is turned out that dvc natively does not support such cases when a parameter is a result of some python code (for example x = pandas.Timestamp.today()). As a result, to refit the model automatically I need whether to code some side custom solution which basically follows the same logic dvc does or to pass -f flag manually, which makes this task non-automatic. So it would be very nice if such feature support is added!

Thank you :)

[efiop]

Hi :) Not running your code was a conscious choice, as dvc is language-agnostic and we don't execute any of user file's at all. If you need to execute some code you could put it into your pipeline as a stage, that would run and create params file that will then be used in the pipeline.

[dberenbaum]

Right, you can workaround this by having an always-changed stage that outputs your parameters.

However, this is not obvious and is an inconvenient workaround. At least, we should probably document this pattern for other users. We should also consider whether we can make it easier to handle more complex parameter scenarios like this one. @VGODIE, I don't think we need to reopen this issue, but there is a related discussion in #6506 if you want to comment there.