A common way to consider rhythmic complexity is to compute it locally by studying the composition of rhythmic patterns. Many complexity measures have been proposed in the literature [Thul2008] and correlated with human judgment [Beltran2015]. But the perception of the rhythmic complexity of an entire piece of music can depend on more global parameters such as the number of rhythmic patterns used over time, the similarity between the different rhythmic patterns used, the repetition of patterns, their periodicity, the level of novelty throughout the piece, the meter, the tempo... We propose here to consider those characteristics to compute rhythmic complexity in symbolic music pieces at several scales: low level (bar wide), middle level (groups of bars) and high level (piece wide). We present first experiments on symbolic music datasets to study the progression of rhythmic complexity throughout the pieces. This approach can be used to select pieces with particular rhythmic properties in existing corpora, for example in a music therapy context or for educational purposes. It could also be applied for music structure analysis or generation of music with given rhythmic profiles. [Thul 2008] Eric Thul. Measuring the Complexity of Musical Rhythm. PhD the- sis, School of Computer Science, McGill University, 2008. [Beltran 2015] Beltran, J. F., Liu, X., Mohanchandra, N., & Toussaint, G. T. Measuring musical rhythm similarity: Statistical features versus transformation methods. International Journal of Pattern Recognition and Artificial Intelligence, 29(02), 2015.