Memory Efficient Deconstructed Vectorized Dataframe Interface.
Design goals:
- Favor performance over nice syntax features. Sacrifice fool-proof for efficient zero-copy operations.
- Ensure ideal micro-performance and optimize for moderate data sizes (megabytes).
- The use-case is API server code that you write once and execute many times.
- Try to stay compatible with the Pandas interface. There is no
Series
, however. - Rely on numpy.
- Friends with Arrow.
- Frequently release GIL and depend on native extensions doing unsafe things.
- Test only CPython and Linux.
- Support only x86-64 CPUs with AVX2.
- Support only Python 3.10+.
- 100% test coverage.
Otherwise, you should be way better with regular Pandas.
Medvedi is currently heavily used in production of Athenian.