Just to play?
First of all, do you have the equipment to see what you're doing? A scope with 1GHz bandwidth is an obvious starting point; if you hope to see any harmonics, 5G, 10G or more would be handy. Big $$$, but the most familiar relative to time-domain slow stuff.
Specturm analyzer is much cheaper, even out past 10G or so, but doesn't show time domain activity as well. To do that, you need to infer things from sample period (e.g., 60Hz ripple modulating an amplifier, causing amplitude modulation, only noticeable at certain refresh rates), sidebands and so on. Everything's sideways-upside-down in the frequency domain. But it's quite helpful for radio-ey things, where sidebands and harmonics are important and easy to pick out.
You can plod along with, say, an RF voltage probe, but you don't know anything about what's happening, just that it's there or not.
Second, you need parts to do it. MMICs are one thing. BJTs come in that range, as do [Si] JFETs, MOS and the fancier PHEMTs and stuff. Offhand, I don't have any stuff that goes that high, other than parts I've salvaged that I can't find full datasheets for, if anything at all. Some good starting points are BFQxxx's and such, many of which have been mentioned here before; others can volunteer their favorite HF jellybeans.
Finally, you need the understanding. Have you played with radio style circuits before, and got a grasp of impedance matching and transformation and inductors and capacitors and filter design and...? It's easy to see the analogy between an inductor and a thin trace, and a capacitor and a thick trace; it's only another step to build a GHz oscillator with 1/4 wave stubs, or a filter with various segments stacked up. Of course, it's another to design one from first principles and have it work correctly (desired frequency, bandwidth and ripple, etc.), and it would be a whole thing further to actually build one from copper clad with an X-Acto blade! But as with lumped filters, there are design tools for that (ranging from simple calculators to free programs to kilobuck design suites). So if you already have a feeling for this sort of stuff, go for it. If you don't, you might plan some more projects in the VHF range, where inductors and capacitors are still reasonably inductive and capacitive, respectively, keeping in mind your experience is directly applicable.
Some projects are pretty simple, too, even with whatever considerations. A pulsed oscillator and wideband detector can be used for very crude radar, and it doesn't really even matter what frequency the oscillator runs at. If it's making an RF signal, you can be reasonably assured it's doing it at whatever GHz your stubs and stuff are tuned for. Zap it off an antenna and watch the return blip. Stuff like this was done back in the days when scopes were barely pushing 10MHz, and GHz was generated with freaky vacuum tubes, like klystrons, backward-wave oscillators and magnetrons.
Tim