sketch is a generic, header-only library providing implementations of a variety of sketch data structures for scalable and streaming applications.
All have been accelerated with SIMD parallelism where possible, most are composable, and many are threadsafe unless -DNOT_THREADSAFE
is passed as a compilation flag.
- HyperLogLog Implementation [hll.h]
hll_t
/hllbase_t<HashStruct>
- Estimates the cardinality of a set using log(log(cardinality)) bits.
- Threadsafe unless
-DNOT_THREADSAFE
is passed. - Currently,
hll
is the only structure for which python bindings are available, but we intend to extend this in the future.
- HyperBitBit [hbb.h]
- Better per-bit accuracy than HyperLogLogs, but, at least currently, limited to 128 bits/16 bytes in sketch size.
- Bloom Filter [bf.h]
bf_t
/bfbase_t<HashStruct>
- Naive bloom filter
- Currently not threadsafe.
- Count-Min and Count Sketches
- ccm.h (
ccmbase_t<UpdatePolicy=Increment>/ccm_t
(usepccm_t
for Approximate Counting orcs_t
for a count sketch). - The Count sketch is threadsafe if
-DNOT_THREADSAFE
is not passed or if an atomic container is used. Count-Min sketches are currently not threadsafe due to the use of minimal updates. - Count-min sketches can support concept drift if
realccm_t
from mult.h is used.
- ccm.h (
- MinHash sketches
- mh.h (
RangeMinHash
is the currently verified implementation.) We recommend you build the sketch and then convert to a linear container (e.g., astd::vector
) usingto_container<ContainerType>()
or.finalize()
for faster comparisons. - CountingRangeMinHash performs the same operations as RangeMinHash, but provides multiplicities, which facilitates
histogram_similarity
, a generalization of Jaccard with multiplicities. - Both CountingRangeMinHash and RangeMinHash can be finalized into containers for fast comparisons with
.finalize()
. - A draft HyperMinHash implementation is available as well, but it has not been thoroughly vetted.
- Range MinHash implementations and the HyperMinHash implementation are not threadsafe.
- mh.h (
- B-Bit MinHash
- bbmh.h
- One-permutation (partition) bbit minhash
- Threadsafe, bit-packed and fully SIMD-accelerated
- Power of two partitions are supported in BBitMinHasher, which is finalized into a FinalBBitMinHash sketch. This is faster than the alternative.
- We also support arbitrary divisions using fastmod64 with DivBBitMinHasher and its corresponding final sketch, FinalDivBBitMinHash.
- One-permutation counting bbit minhash
- Not threadsafe.
- HeavyKeeper
- hk.h
- Not threadsafe currently.
- Reference: https://www.usenix.org/conference/atc18/presentation/gong
- A seemingly unilateral improvement over count-min sketches.
- One drawback is the inability to delete items, which makes it unsuitable for sliding windows.
- It shares this characteristic with the Count-Min sketch with conservative update and the Count-Min Mean sketch.
- ntcard
- mult.h
- Threadsafe
- Reference: https://www.ncbi.nlm.nih.gov/pubmed/28453674
- Not SIMD-accelerated, but also general, supporting any arbitrary coverage level
The following sketches are experimental or variations on prior structures
- HyperLogFilter [hll.h]
hlf_t
/hlfbase_t<HashStruct>
,chlf_t
/chlfbase_t<HashStruct>
- New data structure which provides the same quantitative accuracy as a HyperLogLog while providing more effective approximate membership query functionality than the HyperLogLog.
chlf_t
is identical to thehlf_t
structure, with the exception that the memory is contiguous and each sketch cannot be used individually.- Threadsafe unless
-DNOT_THREADSAFE
is passed.
- filterhll [filterhll.h]
fhll_t
/fhllbase_t<HashStruct>
- Simple hll/bf combination without rigorous guarantees for requiring an element be present in the bloom filter to be inserted into the HyperLogLog.
- Currently not threadsafe.
- Naive Approximate Counting Bloom Filter [cbf.h]
cbf_t
/cbfbase_t<HashStruct>
- An array of bloom filters where presence in a sketch at a given index replaces the count for the approximate counting algorithm.
- Currently not threadsafe.
- Probabilistic Counting Bloom Filter
pcbf_t
/pcbfbase_t<HashStruct>
- An array each of bloom filters and hyperloglogs for approximate counting. The hyperloglogs provide estimated cardinalities for inserted elements, which allows us to estimate the error rates of the bloom filters and therefore account for them in count estimation The hyperloglogs provide estimated cardinalities for inserted elements, which allows us to estimate the error rates of the bloom filters and therefore account for them in count estimation.
- Currently not threadsafe.
- Multiplicity Adapter
- In
mult.h
, the WeightedSketch template class uses a point querying counting structure (by default, a count-min sketch, but we have observed better results with HeavyKeeper). - You can find an example of an adapter providing a weighted jaccard HyperLogLog in
test/multtest.cpp
. This works for any of the structures working on unweighted sets.
- In
Future work
- Multiplicities
- Consistent Weighted Sampling, Improved CWS
- Sampling algorithms and core-sets
To build and run the hll test case:
make test && ./test
To use:
using namespace sketch;
hll::hll_t hll(14); // Use 2**14 bytes for this structure
// Add hashed values for each element to the structure.
for(uint64_t i(0); i < 10000000ull; ++i) hll.addh(i);
fprintf(stderr, "Elements estimated: %lf. Error bounds: %lf.\n", hll.report(), hll.est_err());
The other structures work with a similar interface. See the type constructors for more information or view 10xdash for examples on using the same interface for a variety of data structures.
Simply #include sketch/<header_name>
, or, for one include #include <sketch/sketch.h>
, which places all subnamespaces in the sketch namespace,
which allows you to write sketch::bf_t
and sketch::hll_t
without the subnamespaces.
By default, updates to the hyperloglog structure to occur using atomic operations, though threading should be handled by the calling code. Otherwise, the flag -DNOT_THREADSAFE
should be passed. The cost of this is relatively minor, but in single-threaded situations, this would be preferred.
Python bindings are available via pybind11. Simply cd python && python setup.py install
.