fatal: Split Bloom Filter bf() should be SBF.

[membranepotential]

Hello,

Unfortunately, bt build does not work currently. Apparently, dynamic casting of a BF* to an SBF* fails. I tried to find error sources, but was unable to narrow it down.

Maybe it is connected with getting rid of the overloaded functions SBF::load()?

Cheers!

[Bradsol]

Is this an error in the release version or the most recent commit (or some specific commit prior to the latest)?

[Bradsol]

Additionally can you provide the command you used to build a bloom filter?

[membranepotential]

The error appeared with the latest commit and

# Create hash function
$ ~/packages/splitsbt/bt hashes ssbt_hashes 1

# Make bloomfilters with
$ BF_SIZE=2000000000
$ ~/packages/splitsbt/bt count --threads 4 ssbt_hashes.bin $BF_SIZE <(gunzip -c raw/$SRA/*) "bloomfilters/${SRA}.bf.bv"

# Make file listing the bloomfilters
$ head -n1 bloomfilter_list.txt
/home/sra_index/bloomfilters/DRR008828.bf.bv

# Build the SSBT
$ ~/packages/splitsbt/bt build ssbt_hashes.bin bloomfilter_list.txt ssbt.bloomtree
Starting Bloom Tree
Kmer size = 20
Building...
With build type 2
Building tree on 1763 bloom filters.
Loading hashes from ssbt_hashes.bin
# Hash applications=1
Read hashes for k=20
Inserting leaves into tree...
Inserting leaf /home/sra_index/bloomfilters/DRR008828.bf.bv ...
Inserting leaf /home/sra_index/bloomfilters/DRR015133.bf.bv ...
At node: /home/sra_index/bloomfilters/DRR008828.bf.bv
Load_size: 2000000000
Load_size: 2000000000
fatal: Split Bloom Filter bf() should be SBF.

I couln't find a release version, which commit would you suggest I try?

Thank you!

[membranepotential]

I now tried again at commit a09178588b7f82a9d0e922ef4fdb3cc466a606dd and get the same error.

[Bradsol]

Ah you are attempting to build an SSBT using SBT filters. If you rename your filters to the extension '.sim.bf.bv' it should work fine. For example DRR008828.sim.bf.bv instead of DRR008828.bf.bv

As a fun fact, the current commit actually has a new build algorithm which I'm in the process of organizing into a release. It's significantly faster to build and produces a better tree but requires constructing minhash sketches of the input bit vectors. I'll see if I can write an updated user manual for it by the weeks end.

[membranepotential]

Oh wow that was much easier to fix than I thought :-)

So would you expect that bt build works with the current commit? I see the bt minhash command in the source, but bt build seems unchanged in main.cc.

Thanks!

[Bradsol]

Yes - the development of the new construction method is orthogonal to the existing build and doesn't use any of the same code. In theory (though I haven't rigorously tested it) the old build should be unchanged.