hahaXD / structured_bn

Structured Bayesian Networks

This project aims to compile a structured Bayesian network that is described by a json file to a single PSDD. For example, medium_sf_sdd.json is an example of a json file that describes an SBN. Further, the direcotry medium_sf_sdds contains many sdd files that are referenced from the json file.

Compile

mkdir build
cd build
cmake3 ..
make

After running the commands above, two binaries should be generated, structured_bn_main and structured_bn_test. To test the setup, please run

./structured_bn_test

You should see all Passes.

Run

./structured_bn_main --learning_dataset=<path_to_dataset_file> --psdd_filename <output_psdd_filename> --vtree_filename <output_vtree_filename> <sbn_json_file> 

This will first learn weights of a SBN structure that is specified by <sbn_json_file>, and compiled the learned SBN into a single PSDD. The compiled PSDD is stored in output_psdd_filename and the used vtree is also stored in output_vtree_filename.

For example, the following command compiles an SBN whose structure is specified by medium_sf_sdd.json. The weights of the SBN is randomly sampled, instead of being learned from a dataset.

./structured_bn_main --sample_parameter --psdd_filename=medium_sf_sdd.psdd --vtree_filename=medium_sf_sdd.vtree  medium_sf_sdd.json

Dataset

The learning dataset is a comma separated file, where the column i represents the value of variable i+1. The index of a variable starts from 1.

For example, the following datafile specifies ten training examples, five of each valid instantiation. One valid instantiation is where variable 1 is 1 and variable 2 is 0. The other valid instantiation is where variable 1 is 0 and variable 2 is 1.

1,0
0,1
0,1
1,0
0,1
1,0
1,0
0,1
1,0
0,1