private double distance(Metadata m1, Metadata m2) {
double wAttribute_type = 1;
double wDefault_task = 1;
double wMissing_values = 1;
double wNumber_of_attributes = 1;
double wNumber_of_instances = 1;
// can be null
double dAttribute_type = dCategory(m1.getAttribute_type(), m2.getAttribute_type());
double dDefault_task = dCategory(m1.getDefault_task(), m2.getDefault_task());
// default false - always set
double dMissing_values = dBoolean(m1.getMissing_values(), m2.getMissing_values());
// mandatory attributes - always set
double dNumber_of_attributes =
d(m1.getNumber_of_attributes(), m2.getNumber_of_attributes(), minAttributes, maxAttributes);
double dNumber_of_instances =
d(m1.getNumber_of_instances(), m2.getNumber_of_instances(), minInstances, maxInstances);
double distance =
wAttribute_type * dAttribute_type
+ wDefault_task * dDefault_task
+ wMissing_values * dMissing_values
+ wNumber_of_attributes * dNumber_of_attributes
+ wNumber_of_instances * dNumber_of_instances;
return distance;
}
@Override
protected Agent chooseBestAgent(Data data) {
Metadata metadata = data.getMetadata();
GetAllMetadata gm = new GetAllMetadata();
gm.setResults_required(true);
// 1. choose the nearest training data
List allMetadata = DataManagerService.getAllMetadata(this, gm);
// set the min, max instances and attributes first
Iterator itr = allMetadata.iterator();
while (itr.hasNext()) {
Metadata next_md = (Metadata) itr.next();
int na = next_md.getNumber_of_attributes();
minAttributes = Math.min(minAttributes, na);
maxAttributes = Math.max(maxAttributes, na);
int ni = next_md.getNumber_of_instances();
minInstances = Math.min(ni, minInstances);
maxInstances = Math.max(ni, maxInstances);
}
ArrayList<MetadataDistancePair> distances = new ArrayList<MetadataDistancePair>();
itr = allMetadata.iterator();
while (itr.hasNext()) {
Metadata next_md = (Metadata) itr.next();
double dNew = distance(metadata, next_md);
distances.add(new MetadataDistancePair(next_md, dNew));
}
Collections.sort(distances);
List agents = new LinkedList();
for (int i = 0; i < M; i++) {
log(distances.get(i).m.getExternal_name() + ": " + distances.get(i).d);
List ag = DataManagerService.getTheBestAgents(this, distances.get(i).m.getInternal_name(), N);
Iterator it = ag.iterator();
while (it.hasNext()) {
agents.add(it.next());
}
}
HashMap<String, Integer> counts = new HashMap<String, Integer>();
Iterator it = agents.iterator();
while (it.hasNext()) {
Agent a = (Agent) it.next();
if (counts.containsKey(a.getType())) {
counts.put(a.getType(), counts.get(a.getType()) + 1);
} else {
counts.put(a.getType(), 1);
}
}
int maxCount = 0;
String bestAgentType = null;
for (String s : counts.keySet()) {
log(s + ": " + counts.get(s));
if (counts.get(s) > maxCount) {
maxCount = counts.get(s);
bestAgentType = s;
}
}
log("Best agent: " + bestAgentType);
ArrayList<Agent> bestAgentOptions = new ArrayList<Agent>();
it = agents.iterator();
while (it.hasNext()) {
Agent a = (Agent) it.next();
if (a.getType().equals(bestAgentType)) {
bestAgentOptions.add(a);
}
}
List optionSamples = getAgentOptions(bestAgentType);
List options = new LinkedList();
it = optionSamples.iterator();
while (it.hasNext()) {
Option o = (Option) it.next();
Option newOpt = o.copyOption();
// ignore boolean and set options for now, set their value to the one of the best agent on
// closest file
if (o.getData_type().equals("BOOLEAN") || o.getData_type().equals("MIXED")) {
if (bestAgentOptions.get(0).getOptionByName(o.getName()) == null) {
continue;
}
newOpt.setValue(bestAgentOptions.get(0).getOptionByName(o.getName()).getValue());
} else {
double sum = 0;
int count = 0;
String optionName = o.getName();
for (Agent a : bestAgentOptions) {
if (a.getOptionByName(optionName) != null) {
sum += Double.parseDouble(a.getOptionByName(optionName).getValue());
}
count++;
}
double avg = sum / count;
double stdDev = 0;
for (Agent a : bestAgentOptions) {
if (a.getOptionByName(optionName) != null) {
stdDev +=
Math.pow(Double.parseDouble(a.getOptionByName(optionName).getValue()) - avg, 2);
}
}
stdDev = Math.sqrt(stdDev / count);
if (stdDev > 0) {
newOpt.setValue("?");
newOpt.setUser_value("?");
newOpt.setMutable(true);
Interval range = new Interval();
range.setMin((float) Math.max(avg - 2 * stdDev, o.getRange().getMin()));
range.setMax((float) Math.min(avg + 2 * stdDev, o.getRange().getMax()));
newOpt.setRange(range);
} else {
if (o.getData_type().equals("FLOAT")) {
newOpt.setValue(Double.toString(avg));
}
if (o.getData_type().equals("INT")) {
newOpt.setValue(Integer.toString((int) avg));
}
}
}
options.add(newOpt);
}
Agent agent = new Agent();
agent.setName(null);
agent.setType(bestAgentType);
agent.setOptions(options);
return agent;
}
