I like Human Ressource Machine, because it rates me on code length and complexity. But the types of algorithms seems limited.
I like project Euler for its variety, but I don't get any feedback on quality of my solutions.
Is there a good middle ground?
They have an automated grader that runs through a number of test cases (some of which are hidden from you). Unfortunately, after the first few problems unless you're a paid student you only get an input file and a box to enter your result in.
No direct feedback on complexity, but the automated grader has some limits on execution time and memory use that at least tell you whether your implementation is "good enough."