public static ICModel runSparsifier(
SocialNetwork socNet,
ObservationsReader observations,
ICModel originalModel,
int sparseSize,
Sparsifier sparse,
int numOfChunks,
boolean reportPartial) {
LOGGER.info(
"Arcs: total in social network="
+ socNet.getArcs().size()
+ ", with non-zero probability="
+ originalModel.getProbs().cardinality()
+ ", target k="
+ sparseSize);
LOGGER.info("Begin creating sparsified model");
ICModel sparseModel = sparse.sparsify(sparseSize, numOfChunks, observations, reportPartial);
LOGGER.info("Done creating sparsified model");
LOGGER.info("Begin measuring the resulting sparsified model");
// Report change in log likelihood
double sparsifiedLogL =
sparseModel.getLogLikelihoodIgnoringParentInformation(sparse.getAuxiliary());
LOGGER.info("SPARSIFIED log likelihood (ignoring parent information)=" + sparsifiedLogL);
double originalLogL =
originalModel.getLogLikelihoodIgnoringParentInformation(sparse.getAuxiliary());
LOGGER.info("ORIGINAL log likelihood (ignoring parent information)=" + originalLogL);
LOGGER.info("End measuring the resulting sparsified model");
return sparseModel;
}
public static void main(String[] args) throws Exception {
final SimpleJSAP jsap =
new SimpleJSAP(
Sparsifier.class.getName(),
"Estimates and sparsifies a propagation model from a set of observations.",
new Parameter[] {
new FlaggedOption(
"social-network",
JSAP.STRING_PARSER,
JSAP.NO_DEFAULT,
JSAP.REQUIRED,
's',
"social-network",
"The file containing the social network graph"),
new FlaggedOption(
"probabilities",
JSAP.STRING_PARSER,
JSAP.NO_DEFAULT,
JSAP.REQUIRED,
'p',
"probabilities",
"The file containing the propagation probabilities"),
new FlaggedOption(
"candidate-selection-policy",
JSAP.STRING_PARSER,
CandidateSelectionPolicy.DEFAULT_CANDIDATE_SELECTION_POLICY
.getClass()
.getSimpleName(),
JSAP.REQUIRED,
'c',
"candidate-selection-policy",
"The name of the candidate selection policy"),
new FlaggedOption(
"auxiliary-basename",
JSAP.STRING_PARSER,
JSAP.NO_DEFAULT,
JSAP.NOT_REQUIRED,
JSAP.NO_SHORTFLAG,
"auxiliary-basename",
"The base name for reading a pre-computed auxiliary structure"),
new FlaggedOption(
"input",
JSAP.STRING_PARSER,
JSAP.NO_DEFAULT,
JSAP.REQUIRED,
'i',
"input",
"The file containing the observations"),
new FlaggedOption(
"sparsifier",
JSAP.STRING_PARSER,
DEFAULT_SPARSIFIER.getSimpleName(),
JSAP.NOT_REQUIRED,
'f',
"sparsifier",
"The sparsifier to run, from this list: "
+ StringUtils.join(Reflection.subClasses(Sparsifier.class), ',')),
new FlaggedOption(
"sparse-model-size",
JSAP.INTEGER_PARSER,
JSAP.NO_DEFAULT,
JSAP.NOT_REQUIRED,
'k',
"sparse-model-size",
"The size of the sparse model"),
new FlaggedOption(
"number-of-chunks",
JSAP.INTEGER_PARSER,
JSAP.NO_DEFAULT,
JSAP.NOT_REQUIRED,
'r',
"number-of-chunks",
"The number of chunks to be sparsified in parralel"),
new FlaggedOption(
"output",
JSAP.STRING_PARSER,
JSAP.NO_DEFAULT,
JSAP.NOT_REQUIRED,
'o',
"output",
"File to dump sparsified model to"),
new FlaggedOption(
"measures-file",
JSAP.STRING_PARSER,
JSAP.NO_DEFAULT,
JSAP.NOT_REQUIRED,
'z',
"measures-file",
"Save measures of partial models to file"),
new FlaggedOption(
"debug-file",
JSAP.STRING_PARSER,
JSAP.NO_DEFAULT,
JSAP.NOT_REQUIRED,
'd',
"debug-file",
"Save debug information to file"),
new Switch(
"with-fraction",
'n',
"with-fraction",
"Disable the computation of the 'fraction of covered propagations'."),
new Switch(
"incremental-likelihood",
JSAP.NO_SHORTFLAG,
"incremental-likelihood",
"Performs incremental computation of likelihood, for sparsifications methods that support this option (faster, experimental)."),
});
final JSAPResult jsapResult = jsap.parse(args);
if (jsap.messagePrinted()) {
return;
}
// Load social network and input
String snFilename = jsapResult.getString("social-network");
SocialNetwork socNet = new SocialNetwork(Utilities.getIterator(snFilename));
String obsFilename = jsapResult.getString("input");
ObservationsReader observations = new ObservationsReader(obsFilename);
// Load original model
ICModel originalModel =
new ICModel(socNet, Utilities.getIterator(jsapResult.getString("probabilities")));
// Load candidate selection policy
String selectionPolicyName = jsapResult.getString("candidate-selection-policy");
Class<?> selectionPolicyClass =
Class.forName(
CandidateSelectionPolicy.class.getPackage().getName() + "." + selectionPolicyName);
CandidateSelectionPolicy candidateSelectionPolicy =
(CandidateSelectionPolicy)
selectionPolicyClass.getConstructor(new Class[] {}).newInstance(new Object[] {});
// Create sparsifier
String sparsifierName = jsapResult.getString("sparsifier");
Class<?> sparsifierClass =
Class.forName(Sparsifier.class.getPackage().getName() + "." + sparsifierName);
Sparsifier sparsifier =
(Sparsifier)
sparsifierClass
.getConstructor(new Class[] {originalModel.getClass()})
.newInstance(new Object[] {originalModel});
LOGGER.info("Created a " + sparsifier.getClass().getSimpleName());
// Set sparsifier options
if (!jsapResult.getBoolean("with-fraction")) {
sparsifier.disableComputationOfPartialFractionOfPropagations();
LOGGER.info("Disabled the computation of fraction of propagations");
}
if (jsapResult.getBoolean("incremental-likelihood")) {
if (sparsifier instanceof GreedySparsifier) {
((GreedySparsifier) sparsifier).setIncrementalLikelihoodComputation();
LOGGER.info("Enabled incremental computation of likelihood (faster, experimental)");
} else {
LOGGER.warn(
"This type of sparsifier does not accept the --incrementa-likelihood switch, ignoring");
}
}
if (jsapResult.userSpecified("auxiliary-basename")) {
// Use existing auxiliary file
String auxiliaryBasename = jsapResult.getString("auxiliary-basename");
ICEstimateAuxiliary auxiliary = new ICEstimateAuxiliary(socNet, observations, null);
LOGGER.info("Loading pre-computed auxiliary variables");
auxiliary.read(auxiliaryBasename);
sparsifier.useAuxiliary(auxiliary);
if (auxiliary
.getCandidateSelectionPolicy()
.toSpec()
.equals(candidateSelectionPolicy.toSpec())) {
LOGGER.info(
"Candidate selection policy: " + auxiliary.getCandidateSelectionPolicy().toSpec());
} else {
throw new IllegalArgumentException(
"The candidate selection policies do not match: auxiliary has '"
+ auxiliary.getCandidateSelectionPolicy().toSpec()
+ "', sparsifier has '"
+ candidateSelectionPolicy.toSpec()
+ "'");
}
} else {
// Compute auxiliary variables
LOGGER.info("Computing auxiliary variables");
sparsifier.computeAuxiliary(observations, candidateSelectionPolicy);
}
int maxSparseSize;
if (jsapResult.userSpecified("sparse-model-size")) {
maxSparseSize = jsapResult.getInt("sparse-model-size");
} else {
maxSparseSize = originalModel.getProbs().cardinality();
LOGGER.info(
"Setting target number of arcs to number of arcs with non-zero probability in the original model");
}
// Open debug file
if (jsapResult.userSpecified("debug-file")) {
String debugFilename = jsapResult.getString("debug-file");
LOGGER.info("Will write debug output to " + debugFilename);
sparsifier.openDebugFile(debugFilename);
}
int numOfChunks = 1;
if (jsapResult.userSpecified("number-of-chunks")) {
numOfChunks = jsapResult.getInt("number-of-chunks");
}
ICModel sparseModel =
runSparsifier(
socNet, observations, originalModel, maxSparseSize, sparsifier, numOfChunks, true);
// Write partial results to file if necessary
if (jsapResult.userSpecified("measures-file")) {
for (Measure m : sparsifier.partialResults.keySet()) {
String logFilename = jsapResult.getString("measures-file");
switch (m) {
case LOG_L:
logFilename = logFilename + ".logL";
break;
case FRACTION_OF_PROPAGATIONS:
logFilename = logFilename + ".frac";
break;
default:
break;
}
PrintWriter report =
new PrintWriter(new BufferedWriter(new FileWriter(new File(logFilename))));
LOGGER.info("Writing partial " + m.toString() + " results to " + logFilename);
report.println("#k\t" + m.toString());
int[] ks = sparsifier.partialResults.get(m).keySet().toArray(new int[] {});
Arrays.sort(ks);
for (int k : ks) {
report.println(k + "\t" + sparsifier.partialResults.get(m).get(k));
}
report.close();
}
}
// Dump probabilities
if (jsapResult.userSpecified("output")) {
String probsFilename = jsapResult.getString("output");
PrintWriter pw = Utilities.getPW(probsFilename);
LOGGER.info("Dumping probabilities to " + probsFilename);
sparseModel.dumpProbabilities(pw);
pw.close();
}
sparsifier.closeDebugFile();
}