The large available spectrum at millimeter wave (mmWave) frequencies can enable the gigabit-per- second data rates needed for next generation wireless systems. To compensate for the high propagation loss at mmWave, a large number of antennas are usually employed to enable beamforming. However, the frequency spectrum can be more efficiently utilized through spatial multiplexing. In conventional multi-antenna systems implemented at lower frequencies, spatial multiplexing is realized by baseband digital precoding. Considering the high power consumption of mixed-signal devices at mmWave, a hybrid analog/digital precoding structure is more attractive, where a large number of antennas share only a few RF chains. This structure limits the flexibility in precoding/combining, because the options for the analog precoder/combiner are usually constrained by the finite-size of the codebooks. In this paper, we propose a new metric, singular space coverage, for selecting analog precoding/combining vectors from the predefined codebooks. We propose three algorithms to select analog precoding/combining vectors that maximize this metric. Compared with the orthogonal matching pursuit based algorithms, our algorithms achieve higher spectral efficiency when using smaller-size codebooks, which are of great interest for simplifying the system structure and reducing the beamsteering training overhead.