@misc{bercea2019fullydynamic,
    title={Fully-Dynamic Space-Efficient Dictionaries and Filters with Constant Number of Memory Accesses},
    author={Ioana O. Bercea and Guy Even},
    year={2019},
    eprint={1911.05060},
    archivePrefix={arXiv},
    primaryClass={cs.DS}
}

See https://arxiv.org/pdf/1911.05060.pdf page 12, for pd single page explanation on Pocket Dictionary, that is self-contained, and includes drawings.

git clone https://github.com/TomerEven/Pocket_Dictionary.git
cd Pocket_Dictionary/Filter_PD
mkdir build_dir
cd build_dir
cmake ..
make

In build_dir directory

• There are 3 types of benchmarking for PD based filter:

./Filter_PD <type>

./Filter_PD <type> <max number of elements> <error probability exponent> <number of lookup to preform>

./Filter_PD <type> <max number of elements> <error probability exponent> <number of lookup to preform> <first level counter size> <second level counter size> <first level load factor> <second level load factor> 

• type: 0 or 1.
  0 is for pd filter based on Pocket Dictionary.
  1 is for pd filter based on counting PD (CPD)

• max number of elements: The max number of distinct elements the filter is supposed to contain in any time. This effects the space usage.

• error probability exponent: Set the probability the filter will err (False positive), to be 2^error probability exponent

• number of lookup to preform: Number of random lookups that will preformed.

• first level counter size: Number of bits each counter in the CPD has.

• second level counter size: Number of bits each counter in the counter hash table. Also maximal number of repetition each element can have.

• first level load factor: In average, how loaded should each PD be.

• second level load factor: Sets the hash table size.

Comparing filters for benchmarks.