[sophacles]

I don't know that #1 was a mistake. When getting into any new field, it is very hard to tell what is important and what isn't, what is a solution to the problem and what isn't. If you are in school, sure they'll guide you, but if you are self-learning, there is no really good filter. So learn about everything. Take a breadth-first approach. Some of the specifics you learn will be wrong, useless, or otherwise inappropriate. However, getting an understanding as to why they aren't relevant is in fact learning the field.

The more information about a topic you have in your head, the more likely you are to see the connections between subfields, how various bits are important and unique or just ho-hum standard implementations.

I regularly see people in some niche area say "look we have found this amazing new way to handle the data-flood", only to tell them they have reinvented NoSQL or a message queue, or a technique for dealing with matrices that the image processing or simulation folks have been doing for years (and all of them poorly, making recognizable mistakes). A breadth-first overview has much advantage.

[xster]

Agreed. It's easy to look back and know what were needed in retrospect but it's hard to even know which is for what until some time is spent to learn its capabilities