since spark uses lazy execution, those dataframes you do not need cannot be cleared unless you do use them (why define them otherwise?).
So when doing an action, spark will execute all code that is necessary.  If you run into memory issues, you can do several things:
- see if you can write the code differently f.e. using the spark runtime instead of plain py functions
- use a larger driver
- persist dataframes (using checkpoints/cache)
Since spark executes code on the JVM, a garbage collect runs to scrap unecessary memory allocations.