Hi Xinran,
Currently the software is on minimal support due the the original author no longer with the lab. Please ensure that you are trying the tagged release on Github (
https://github.com/hoffmangroup/umap/tags) with Python 2.
My current understanding with the procedure with bisulfite conversion is that two runs must be made , one with a reference genome converted with C to T and another with G to A each with their own index. Umap handles the reverse complement conversion of the converted reference sequence since both forward and reverse strands undergo bisulfite treatment at the same time producing somewhat unintuitive reverse strands.
I would highly recommend trying to get the software working for your own needs as-is. We are working towards a better solution in the meantime. If you are still completely stuck, let me know. I'll have to have a discussion to see what we can do to produce future Bisulfite mappability datasets in the meantime.
Hope that helps,
ESric