There has been a lot of drama for me in the last month and a half—both in my life and at work. I’ve been plugging away a bit at my model, but that’s mostly rearranging furniture at this point. As I’ve mentioned several times, I feel that the real challenge is to get going with the actual RL AI that will learn to play checkers well. Moving from the basics that I’ve been focused on to the real project will be a huge effort. I’ll need to build the whole look-ahead architecture which will include minimax, alpha-beta pruning, MCTS and probably other things I haven’t even considered. Then there’s the models themselves which are the reason for all of this.

Needless to say, this whole project is really interesting to me, and having gotten as far as I have, I would like to be able to bring it to some sort of fruition. To date, most of the application-specific techniques are things that I’ve had to teach myself by reading what I could find. If I’m really serious about learning this, though, it would seem to make sense to learn it formally. As it turns out, there is a Udacity nano-degree specifically for reinforcement learning. That would seem like an ideal set of coursework to really be able to fully understand how to implement what I’ll need. Even better, in the syllabus for the final course, “Multi-Agent Reinforcement Learning,” it specifically shows:

CASE STUDY: ALPHAZERO ● Master the skills behind DeepMind’s AlphaZero.

Well how about that. It’s the exact thing that I’ve been studying as a model for what I’ve been trying to do.

But…

There are things that make this nano-degree very different than the specialization and courses I’ve done on via Deeplearning.ai via Coursera. First is the cost: $1436 for “4 Month Access.” That is significantly more expensive on a per-month basis than what Coursera charges. However, that itself seems reasonable given that it seems like you get a lot more 1-1 mentorship and humans grade the assignments.

The thing that concerns me though, is that it is not a fixed price. If I don’t finish in 4 months, I then need to keep paying a (presumably high) monthly fee. In general, the workload they describe seems manageable… under normal circumstances. But things are not normal for me right now, and I’m concerned that completing the coursework in the 4-month timeframe might be challenging.

The other thing that concerns me is the reviews. In general, the reviews for the degree are excellent. But to the extent that there are any less than stellar reviews, they all revolve around the final section which happens to be the one I listed above. Apparently, the supporting materials for this section (and the one before it) leave a lot to be desired. Perhaps, though, the fact that I’m walking into this with some expereince will balance that out. But perhaps not.

I have until November 12 to figure this out.