Global Parameters

• -t: number of threads (default 1)

• -v: verbose, 0 for silence, 1 for key information and 2 for more information (default 1)

• -j: job, see the table below (default 0 for structure learning)

• -m: method, see the table below (default 0 for PC-Stable)

Note:

MLE = maximum likelihood estimation; JT = juntion tree; VE = variable elimination; LBP = loopy belief propagation; PLS = probabilistic logic sampling; LW = likelihood weighting; SIS = self-importance sampling;

Parameters Related to Structure Learning

For the tasks that include structure learning, the following parameters can be specified through command line options.

• -g: group size, used in the implementation of PC-Stable (default 1)

• -a: significance level alpha, used in constraint-based methods like PC-Stable (default 0.05)

• -ss: whether to save the learned structure, 0 for no, 1 for yes (default 1)

Parameters Related to Parameter Learning

For the tasks that include parameter learning, you can use -sp to specify whether to save the learned parameters or not (default 1).

Parameters Related to Approximate Inference

For the tasks that include structure learning, the following parameters can be specified through command line options.

• -q: desired number of samples, used in sampling-based methods including PLS, LW, SIS, SISv1, AIS-BN, EPIS-BN (default 10,000)

• -u: maximum updating times of importance function, used in learning based importance sampling methods including SIS, SISv1, AIS-BN (default 10)

• -l: updating interval, used in learning based importance sampling methods (default 2,500)

• -d: propagation length, used in LBP and EPIS-BN (default 2)

Parameters Related to Other Functionalities

For other functionalities related to PGM, the following parameters can be specified through command line options.

• -q: number of samples to generate, used in sample generator (default 10,000)

• -c: class variable id (default 0)

Parameters for Files

The files of datasets and networks can be provided via command line options. Please put all the files in FastPGM/dataset and provide the relative path.

• -f0: the PGM that inference operates on or that the samples generated from (default alarm/alarm.xml)

• -f1: the reference PGM that serves as the ground truth of structure learning (default alarm/alarm.bif)

• -f2: training set, used in structure learning, learning, or classification (default alarm/alarm_s5000)

• -f3: testing set, used in inference or classification (default alarm/testing_alarm_1k_p20)

• -f4: the reference potential table file that serves as the ground truth of inference (default alarm/alarm_1k_pt)