private void addSegs(List segs, Sequence output) {
int segtype = -1;
int startidx = -1;
for (int j = 0; j < output.size(); j++) {
// System.out.println("addSegs j="+j);
Object tag = output.get(j);
segtype = ArrayUtils.indexOf(segmentTags, tag.toString());
if (segtype > -1) {
// System.out.println("...found segment "+tag);
// A new segment is starting
startidx = j;
while (j < output.size() - 1) {
// System.out.println("...inner addSegs j="+j);
j++;
Object nextTag = output.get(j);
if (!nextTag.equals(tag)) {
j--;
segs.add(new Segment(segtype, startidx, j));
segtype = startidx = -1;
break;
}
}
}
}
// Handle end-of-sequence
if (startidx > -1) {
segs.add(new Segment(segtype, startidx, output.size() - 1));
}
}
/** The longest dep set of the sub-sequence s[0..idx]. */
public static BitSet longestDepSet(Sequence s, int idx) {
if (s.size() == 0) throw new IllegalArgumentException("size must be greater than 0.");
if (idx < 0 || idx >= s.size()) throw new IllegalArgumentException();
assert s.size() <= maxlength;
int max = -1;
int maxidx = -1;
for (int i = 0; i <= idx; i++) {
BitSet set = sets[i];
set.clear();
set.set(i);
lastuse[i] = i;
for (Variable invar : s.getInputs(i)) {
set.or(sets[lastuse[invar.index]]);
lastuse[invar.index] = i;
}
int size = set.cardinality();
if (size > max) {
max = size;
maxidx = i;
}
}
for (int i = 0; i < s.size(); i++) {
// System.out.println("@ " + sets[i]);
}
return sets[maxidx];
}
@Override
public void addSequence(
Sequence candidate,
int candidateIndex,
int depth,
HashTreeNode father,
ArrayList<Sequence> allCandidates) {
candidateIndicesSize++;
if (candidateIndicesSize > candidateIndices.length) {
if (depth < candidate.size() - 1) {
// exchange this leaf node by inner node if it could become inner node
HashTreeInnerNode newInner = new HashTreeInnerNode();
father.replaceNode(candidate.getItem(depth - 1), newInner);
// and adding all sequences and last candidate
for (int i = 0; i < candidateIndices.length; i++) {
newInner.addSequence(
allCandidates.get(candidateIndices[i]),
candidateIndices[i],
depth,
father,
allCandidates);
}
newInner.addSequence(
allCandidates.get(candidateIndex), candidateIndex, depth, father, allCandidates);
} else {
int[] newIndices = new int[candidateIndices.length * 2];
System.arraycopy(candidateIndices, 0, newIndices, 0, candidateIndices.length);
candidateIndices = newIndices;
candidateIndices[candidateIndicesSize - 1] = candidateIndex;
}
} else {
candidateIndices[candidateIndicesSize - 1] = candidateIndex;
}
}
public Instance pipe(Instance carrier) {
Sequence data = (Sequence) carrier.getData();
Sequence target = (Sequence) carrier.getTarget();
if (data.size() != target.size())
throw new IllegalArgumentException(
"Trying to print into SimpleTagger format, where data and target lengths do not match\n"
+ "data.length = "
+ data.size()
+ ", target.length = "
+ target.size());
int N = data.size();
if (data instanceof TokenSequence) {
throw new UnsupportedOperationException("Not yet implemented.");
} else if (data instanceof FeatureVectorSequence) {
FeatureVectorSequence fvs = (FeatureVectorSequence) data;
Alphabet dict = (fvs.size() > 0) ? fvs.getFeatureVector(0).getAlphabet() : null;
for (int i = 0; i < N; i++) {
Object label = target.get(i);
writer.print(label);
FeatureVector fv = fvs.getFeatureVector(i);
for (int loc = 0; loc < fv.numLocations(); loc++) {
writer.print(' ');
String fname = dict.lookupObject(fv.indexAtLocation(loc)).toString();
double value = fv.valueAtLocation(loc);
// if (!Maths.almostEquals(value, 1.0)) {
// throw new IllegalArgumentException ("Printing to SimpleTagger format: FeatureVector
// not binary at time slice "+i+" fv:"+fv);
// }
writer.print(fname + String.valueOf(value));
}
writer.println();
}
} else {
throw new IllegalArgumentException("Don't know how to print data of type " + data);
}
writer.println();
// writer.print(getDataAlphabet());
return carrier;
}
public ConfidenceEvaluator(InstanceWithConfidence[] instances, boolean sorted) {
this.confidences = new Vector();
for (int i = 0; i < instances.length; i++) {
Sequence input = (Sequence) instances[i].getInstance().getData();
confidences.add(
new EntityConfidence(
instances[i].getConfidence(), instances[i].correct(), input, 0, input.size() - 1));
}
if (!sorted) Collections.sort(confidences, new ConfidenceComparator());
this.nBins = DEFAULT_NUM_BINS;
this.numCorrect = getNumCorrectEntities();
}
public static Sequence getLongestDepSetSubSequence(Sequence s) {
BitSet indices = longestDepSet(s, s.size() - 1);
int length = indices.length();
assert indices.get(length - 1);
Map<Integer, Integer> newIdx = new LinkedHashMap<Integer, Integer>();
int count = 0;
for (int i = 0; i < length; i++) {
if (!indices.get(i)) continue;
newIdx.put(i, count++);
}
Sequence news = new Sequence();
for (int i = 0; i < length; i++) {
if (!indices.get(i)) continue;
List<Variable> oldins = s.getInputs(i);
List<Variable> newins = new ArrayList<Variable>(oldins.size());
for (Variable oldv : oldins) {
newins.add(news.getVariable(newIdx.get(oldv.index)));
}
news = news.extend(s.getStatementKind(i), newins);
}
return news;
}
// This method is responsible for doing two things:
//
// 1. Selecting at random a collection of sequences that can be used to
// create input values for the given statement, and
//
// 2. Selecting at random valid indices to the above sequence specifying
// the values to be used as input to the statement.
//
// The selected sequences and indices are wrapped in an InputsAndSuccessFlag
// object and returned. If an appropriate collection of sequences and indices
// was not found (e.g. because there are no sequences in the componentManager
// that create values of some type required by the statement), the success flag
// of the returned object is false.
@SuppressWarnings("unchecked")
private InputsAndSuccessFlag selectInputs(StatementKind statement) {
Tracer.trace("selectInputs");
// Variable inputTypes containsthe values required as input to the
// statement given as a parameter to the selectInputs method.
List<Class<?>> inputTypes = statement.getInputTypes();
// The rest of the code in this method will attempt to create
// a sequence that creates at least one value of type T for
// every type T in inputTypes, and thus can be used to create all the
// inputs for the statement.
// We denote this goal sequence as "S". We don't create S explicitly, but
// define it as the concatenation of the following list of sequences.
// In other words, S = sequences[0] + ... + sequences[sequences.size()-1].
// (This representation choice is for efficiency: it is cheaper to perform
// a single concatenation of the subsequences in the end than repeatedly
// extending S.)
List<Sequence> sequences = new ArrayList<Sequence>();
// We store the total size of S in the following variable.
int totStatements = 0;
// The method also returns a list of randomly-selected variables to
// be used as inputs to the statement, represented as indices into S.
// For example, given as statement a method M(T1)/T2 that takes as input
// a value of type T1 and returns a value of type T2, this method might
// return, for example, the sequence
//
// T0 var0 = new T0(); T1 var1 = var0.getT1()"
//
// and the singleton list [0] that represents variable var1. The variable
// indices are stored in the following list. Upon successful completion
// of this method, variables will contain inputTypes.size() variables.
// Note additionally that for every i in variables, 0 <= i < |S|.
List<Integer> variables = new ArrayList<Integer>();
// [Optimization]
// The following two variables are used in the loop below only when
// an alias ratio is present (GenInputsAbstract.alias_ratio != null).
// Their purpose is purely to improve efficiency. For a given loop iteration
// i, "types" contains the types of all variables in S, and "typesToVars"
// maps each type to all variable indices of the given type.
SubTypeSet types = new SubTypeSet(false);
MultiMap<Class<?>, Integer> typesToVars = new MultiMap<Class<?>, Integer>();
for (int i = 0; i < inputTypes.size(); i++) {
Class<?> t = inputTypes.get(i);
// TODO Does this ever happen?
if (!Reflection.isVisible(t)) return new InputsAndSuccessFlag(false, null, null);
// true if statement st represents an instance method, and we are currently
// selecting a value to act as the receiver for the method.
boolean isReceiver =
(i == 0 && (statement instanceof RMethod) && (!((RMethod) statement).isStatic()));
// If alias ratio is given, attempt with some probability to use a variable already in S.
if (GenInputsAbstract.alias_ratio != 0
&& Randomness.weighedCoinFlip(GenInputsAbstract.alias_ratio)) {
Tracer.trace("alias_ratio@selectInputs");
// candidateVars will store the indices that can serve as input to the i-th input in st.
List<SimpleList<Integer>> candidateVars = new ArrayList<SimpleList<Integer>>();
// For each type T in S compatible with inputTypes[i], add all the indices in S of type T.
for (Class<?> match : types.getMatches(t)) {
// Sanity check: the domain of typesToVars contains all the types in variable types.
assert typesToVars.keySet().contains(match);
candidateVars.add(
new ArrayListSimpleList<Integer>(
new ArrayList<Integer>(typesToVars.getValues(match))));
}
// If any type-compatible variables found, pick one at random as the i-th input to st.
SimpleList<Integer> candidateVars2 = new ListOfLists<Integer>(candidateVars);
if (candidateVars2.size() > 0) {
int randVarIdx = Randomness.nextRandomInt(candidateVars2.size());
Integer randVar = candidateVars2.get(randVarIdx);
variables.add(randVar);
continue;
}
}
Tracer.trace("NO alias_ratio@selectInputs");
// If we got here, it means we will not attempt to use a value already defined in S,
// so we will have to augment S with new statements that yield a value of type inputTypes[i].
// We will do this by assembling a list of candidate sequences n(stored in the list declared
// immediately below) that create one or more values of the appropriate type,
// randomly selecting a single sequence from this list, and appending it to S.
SimpleList<Sequence> l = null;
// We use one of three ways to gather candidate sequences, but the third case below
// is by far the most common.
if (GenInputsAbstract.always_use_ints_as_objects && t.equals(Object.class)) {
Tracer.trace("always_use_ints_as_objects@selectInputs");
// 1. OBSCURE, applicable only for branch-directed generation project. Get all
// sequences that create one or more integer. Applicable only when inputTypes[i]
// is "Object" and always_use_ints_as_objects option is specified.
if (Log.isLoggingOn()) Log.logLine("Integer-as-object heuristic: will use random Integer.");
l = componentManager.getSequencesForType(int.class, false);
} else if (t.isArray()) {
// 2. If T=inputTypes[i] is an array type, ask the component manager for all sequences
// of type T (list l1), but also try to directly build some sequences that create arrays
// (list l2).
SimpleList<Sequence> l1 = componentManager.getSequencesForType(statement, i);
if (Log.isLoggingOn())
Log.logLine("Array creation heuristic: will create helper array of type " + t);
SimpleList<Sequence> l2 = HelperSequenceCreator.createSequence(componentManager, t);
l = new ListOfLists<Sequence>(l1, l2);
} else {
// 3. COMMON CASE: ask the component manager for all sequences that yield the required type.
if (Log.isLoggingOn()) Log.logLine("Will query component set for objects of type" + t);
l = componentManager.getSequencesForType(statement, i);
}
assert l != null;
if (Log.isLoggingOn()) Log.logLine("components: " + l.size());
// If we were not able to find (or create) any sequences of type inputTypes[i], and we are
// allowed the use null values, use null. If we're not allowed, then return with failure.
if (l.size() == 0) {
Tracer.trace("selectinputs-evalforbid");
if (isReceiver || GenInputsAbstract.forbid_null) {
if (!isReceiver) {
if (GenInputsAbstract.forbid_null) {
Tracer.trace("forbid_null@selectinputs-evalforbid");
}
}
if (Log.isLoggingOn())
Log.logLine("forbid-null option is true. Failed to create new sequence.");
return new InputsAndSuccessFlag(false, null, null);
} else {
if (!GenInputsAbstract.forbid_null) {
Tracer.trace("NOT forbid_null@selectinputs-evalforbid");
}
if (Log.isLoggingOn()) Log.logLine("Will use null as " + i + "-th input");
StatementKind st = PrimitiveOrStringOrNullDecl.nullOrZeroDecl(t);
Sequence seq = new Sequence().extend(st, new ArrayList<Variable>());
variables.add(totStatements);
sequences.add(seq);
assert seq.size() == 1;
totStatements++;
// Null is not an interesting value to add to the set of
// possible values to reuse, so we don't update typesToVars or types.
continue;
}
}
// At this point, we have one or more sequences that create non-null values of type
// inputTypes[i].
// However, the user may have requested that we use null values as inputs with some given
// frequency.
// If this is the case, then use null instead with some probability.
Tracer.trace("selectinputs-null-ratio");
if (!isReceiver
&& GenInputsAbstract.null_ratio != 0
&& Randomness.weighedCoinFlip(GenInputsAbstract.null_ratio)) {
Tracer.trace("null_ratio@selectinputs-null-ratio");
if (Log.isLoggingOn())
Log.logLine("null-ratio option given. Randomly decided to use null as input.");
StatementKind st = PrimitiveOrStringOrNullDecl.nullOrZeroDecl(t);
Sequence seq = new Sequence().extend(st, new ArrayList<Variable>());
variables.add(totStatements);
sequences.add(seq);
assert seq.size() == 1;
totStatements++;
continue;
}
Tracer.trace("NOT null_ratio@selectinputs-null-ratio");
// At this point, we have a list of candidate sequences and need to select a
// randomly-chosen sequence from the list.
Sequence chosenSeq = null;
Tracer.trace("selectInputs-smalltests");
if (GenInputsAbstract.small_tests) {
Tracer.trace("small_tests@selectInputs-smalltests");
chosenSeq = Randomness.randomMemberWeighted(l);
} else {
Tracer.trace("NO small_tests@selectInputs-smalltests");
chosenSeq = Randomness.randomMember(l);
}
// Now, find values that satisfy the constraint set.
Match m = Match.COMPATIBLE_TYPE;
// if (i == 0 && statement.isInstanceMethod()) m = Match.EXACT_TYPE;
Variable randomVariable = chosenSeq.randomVariableForTypeLastStatement(t, m);
// We are not done yet: we have chosen a sequence that yields a value of the required
// type inputTypes[i], but there may be more than one such value. Our last random
// selection step is to select from among all possible values.
// if (i == 0 && statement.isInstanceMethod()) m = Match.EXACT_TYPE;
if (randomVariable == null) {
throw new BugInRandoopException("type: " + t + ", sequence: " + chosenSeq);
}
// If we were unlucky and selected a null value as the receiver
// for a method call, return with failure.
if (i == 0
&& (statement instanceof RMethod)
&& (!((RMethod) statement).isStatic())
&& chosenSeq.getCreatingStatement(randomVariable) instanceof PrimitiveOrStringOrNullDecl)
return new InputsAndSuccessFlag(false, null, null);
// [Optimization.] Update optimization-related variables "types" and "typesToVars".
Tracer.trace("selectinputs-alias");
if (GenInputsAbstract.alias_ratio != 0) {
Tracer.trace("alias_ratio@selectinputs-alias");
// Update types and typesToVars.
for (int j = 0; j < chosenSeq.size(); j++) {
StatementKind stk = chosenSeq.getStatementKind(j);
if (stk instanceof PrimitiveOrStringOrNullDecl)
continue; // Prim decl not an interesting candidate for multiple uses.
Class<?> outType = stk.getOutputType();
types.add(outType);
typesToVars.add(outType, totStatements + j);
}
}
Tracer.trace("NOT alias_ratio@selectinputs-alias");
variables.add(totStatements + randomVariable.index);
sequences.add(chosenSeq);
totStatements += chosenSeq.size();
}
return new InputsAndSuccessFlag(true, sequences, variables);
}
/**
* Tries to create and execute a new sequence. If the sequence is new (not already in the
* specified component manager), then it is executed and added to the manager's sequences. If the
* sequence created is already in the manager's sequences, this method has no effect, and returns
* null.
*/
private ExecutableSequence createNewUniqueSequence() {
Tracer.trace("createNewUniqueSequence");
if (Log.isLoggingOn()) Log.logLine("-------------------------------------------");
StatementKind statement = null;
if (this.statements.isEmpty()) return null;
// Select a StatementInfo
statement = Randomness.randomMember(this.statements);
if (Log.isLoggingOn()) Log.logLine("Selected statement: " + statement.toString());
// jhp: add flags here
InputsAndSuccessFlag sequences = selectInputs(statement);
if (!sequences.success) {
if (Log.isLoggingOn()) Log.logLine("Failed to find inputs for statement.");
return null;
}
Sequence concatSeq = Sequence.concatenate(sequences.sequences);
// Figure out input variables.
List<Variable> inputs = new ArrayList<Variable>();
for (Integer oneinput : sequences.indices) {
Variable v = concatSeq.getVariable(oneinput);
inputs.add(v);
}
Sequence newSequence = concatSeq.extend(statement, inputs);
// With .5 probability, do a primitive value heuristic.
Tracer.trace("heuristic-uniquesequence");
if (GenInputsAbstract.repeat_heuristic && Randomness.nextRandomInt(10) == 0) {
Tracer.trace("repeat_heuristic@heuristic-uniquesequence");
int times = Randomness.nextRandomInt(100);
newSequence = newSequence.repeatLast(times);
if (Log.isLoggingOn()) Log.log(">>>" + times + newSequence.toCodeString());
}
Tracer.trace("NO_repeat_heuristic@heuristic-uniquesequence");
// If parameterless statement, subsequence inputs
// will all be redundant, so just remove it from list of statements.
if (statement.getInputTypes().size() == 0) {
statements.remove(statement);
}
// If sequence is larger than size limit, try again.
Tracer.trace("evaluating-maxsize");
if (newSequence.size() > GenInputsAbstract.maxsize) {
Tracer.trace(">maxsize@evaluating-maxsize");
if (Log.isLoggingOn())
Log.logLine(
"Sequence discarded because size "
+ newSequence.size()
+ " exceeds maximum allowed size "
+ GenInputsAbstract.maxsize);
return null;
}
Tracer.trace("<maxsize@evaluating-maxsize");
randoopConsistencyTests(newSequence);
if (this.allSequences.contains(newSequence)) {
Tracer.trace("discard existing");
if (Log.isLoggingOn())
Log.logLine("Sequence discarded because the same sequence was previously created.");
return null;
}
this.allSequences.add(newSequence);
for (Sequence s : sequences.sequences) {
s.lastTimeUsed = java.lang.System.currentTimeMillis();
}
randoopConsistencyTest2(newSequence);
if (Log.isLoggingOn()) {
Log.logLine("Successfully created new unique sequence:" + newSequence.toString());
}
// System.out.println("###" + statement.toStringVerbose() + "###" + statement.getClass());
// Keep track of any input sequences that are used in this sequence
// Tests that contain only these sequences are probably redundant
for (Sequence is : sequences.sequences) {
subsumed_sequences.add(is);
}
return new ExecutableSequence(newSequence);
}
public S8Vector(Sequence paramSequence) {
this.data = new byte[paramSequence.size()];
addAll(paramSequence);
}
public static BitSet longestDepSet(Sequence s) {
return longestDepSet(s, s.size() - 1);
}
public void test(
Transducer transducer,
InstanceList data,
String description,
PrintStream viterbiOutputStream) {
int[] ntrue = new int[segmentTags.length];
int[] npred = new int[segmentTags.length];
int[] ncorr = new int[segmentTags.length];
LabelAlphabet dict = (LabelAlphabet) transducer.getInputPipe().getTargetAlphabet();
for (int i = 0; i < data.size(); i++) {
Instance instance = data.getInstance(i);
Sequence input = (Sequence) instance.getData();
Sequence trueOutput = (Sequence) instance.getTarget();
assert (input.size() == trueOutput.size());
Sequence predOutput = transducer.viterbiPath(input).output();
assert (predOutput.size() == trueOutput.size());
List trueSegs = new ArrayList();
List predSegs = new ArrayList();
addSegs(trueSegs, trueOutput);
addSegs(predSegs, predOutput);
// System.out.println("FieldF1Evaluator instance "+instance.getName ());
// printSegs(dict, trueSegs, "True");
// printSegs(dict, predSegs, "Pred");
for (Iterator it = predSegs.iterator(); it.hasNext(); ) {
Segment seg = (Segment) it.next();
npred[seg.tag]++;
if (trueSegs.contains(seg)) {
ncorr[seg.tag]++;
}
}
for (Iterator it = trueSegs.iterator(); it.hasNext(); ) {
Segment seg = (Segment) it.next();
ntrue[seg.tag]++;
}
}
DecimalFormat f = new DecimalFormat("0.####");
logger.info(description + " per-field F1");
for (int tag = 0; tag < segmentTags.length; tag++) {
double precision = ((double) ncorr[tag]) / npred[tag];
double recall = ((double) ncorr[tag]) / ntrue[tag];
double f1 = (2 * precision * recall) / (precision + recall);
Label name = dict.lookupLabel(segmentTags[tag]);
logger.info(
" segments "
+ name
+ " true = "
+ ntrue[tag]
+ " pred = "
+ npred[tag]
+ " correct = "
+ ncorr[tag]);
logger.info(
" precision="
+ f.format(precision)
+ " recall="
+ f.format(recall)
+ " f1="
+ f.format(f1));
}
}
public S16Vector(Sequence sequence) {
data = new short[sequence.size()];
addAll(sequence);
}