I would use Postgres.
It's not particularly likely to eat your data or kill you as you get started, and the ecosystem is relatively unified which means consistent and generally good best practices documentation. Becoming a Postgres expert could take a lifetime as the database offers an astonishing depth of features, but it won't bite you if you just use the basics. Plus the source is quite good if you ever find the urge to dig deep.
A couple of gotchas might bother you. You'll need to implement your own strategy for upserts if you need those. But you do get robust native support for popular data types commonly seen in today's apps: geographic data as well as document data (via HStore or JSON).
The MySQL/MariaDB family is well-proven as well, but with tens of available storage engines, two core forks, and a bit of non-standard SQL magic sprinkled throughout it's harder for a beginner to learn the best practices IMO.
The challenge comes once you exceed what one machine can do and want to move beyond the simple client(s) <-> (possibly replicated) server model. Keep in mind that depending on your project, this might not ever happen to you. Certain types of mid-sized sites like Basecamp and Stack Exchange have gotten away with throwing bigger hardware at their database systems.
If or when you reach the point where you need a distributed database, you have to choose between consistency and availability. I'll let the experts explain some of those concepts by linking to this excellent article by Coda Hale: http://codahale.com/you-cant-sacrifice-partition-tolerance/ . That's when you'll find yourself in the NoSQL world - many NoSQL databases are useful in that they try to trade consistent for available in some way or another. While they generally fail completely to actually do so, as illustrated in Kyle Kingsbury's excellent "Call Me Maybe" series, the largest -scale systems are all based on a "NoSQL" data store in some way or another.