// Linear Prediction in information form as in Ref[2]
Float predict (Linear_invertable_predict_model& f)
{
Linear_predict_byproducts b(f.Fx.size1(),f.q.size());
return predict (f, b);
}
bool KNN::train_(LabelledClassificationData &trainingData,UINT K){
//Clear any previous models
clear();
if( trainingData.getNumSamples() == 0 ){
errorLog << "train(LabelledClassificationData &trainingData) - Training data has zero samples!" << endl;
return false;
}
//Set the dimensionality of the input data
this->K = K;
this->numFeatures = trainingData.getNumDimensions();
this->numClasses = trainingData.getNumClasses();
//TODO: In the future need to build a kdtree from the training data to allow better realtime prediction
this->trainingData = trainingData;
if( useScaling ){
ranges = this->trainingData.getRanges();
this->trainingData.scale(ranges, 0, 1);
}
//Set the class labels
classLabels.resize(numClasses);
for(UINT k=0; k<numClasses; k++){
classLabels[k] = trainingData.getClassTracker()[k].classLabel;
}
//Flag that the algorithm has been trained so we can compute the rejection thresholds
trained = true;
//If null rejection is enabled then compute the null rejection thresholds
if( useNullRejection ){
//Set the null rejection to false so we can compute the values for it (this will be set back to its current value later)
bool tempUseNullRejection = useNullRejection;
useNullRejection = false;
rejectionThresholds.clear();
//Compute the rejection thresholds for each of the K classes
vector< double > counter(numClasses,0);
trainingMu.resize( numClasses, 0 );
trainingSigma.resize( numClasses, 0 );
rejectionThresholds.resize( numClasses, 0 );
//Compute Mu for each of the classes
const unsigned int numTrainingExamples = trainingData.getNumSamples();
vector< IndexedDouble > predictionResults( numTrainingExamples );
for(UINT i=0; i<numTrainingExamples; i++){
predict( trainingData[i].getSample(), K);
UINT classLabelIndex = 0;
for(UINT k=0; k<numClasses; k++){
if( predictedClassLabel == classLabels[k] ){
classLabelIndex = k;
break;
}
}
predictionResults[ i ].index = classLabelIndex;
predictionResults[ i ].value = classDistances[ classLabelIndex ];
trainingMu[ classLabelIndex ] += predictionResults[ i ].value;
counter[ classLabelIndex ]++;
}
for(UINT j=0; j<numClasses; j++){
trainingMu[j] /= counter[j];
}
//Compute Sigma for each of the classes
for(UINT i=0; i<numTrainingExamples; i++){
trainingSigma[predictionResults[i].index] += SQR(predictionResults[i].value - trainingMu[predictionResults[i].index]);
}
for(UINT j=0; j<numClasses; j++){
double count = counter[j];
if( count > 1 ){
trainingSigma[ j ] = sqrt( trainingSigma[j] / (count-1) );
}else{
trainingSigma[ j ] = 1.0;
}
}
//Check to see if any of the mu or sigma values are zero or NaN
bool errorFound = false;
for(UINT j=0; j<numClasses; j++){
if( trainingMu[j] == 0 ){
warningLog << "TrainingMu[ " << j << " ] is zero for a K value of " << K << endl;
}
if( trainingSigma[j] == 0 ){
warningLog << "TrainingSigma[ " << j << " ] is zero for a K value of " << K << endl;
}
if( isnan( trainingMu[j] ) ){
errorLog << "TrainingMu[ " << j << " ] is NAN for a K value of " << K << endl;
errorFound = true;
}
if( isnan( trainingSigma[j] ) ){
errorLog << "TrainingSigma[ " << j << " ] is NAN for a K value of " << K << endl;
errorFound = true;
}
}
if( errorFound ){
trained = false;
return false;
}
//Recompute the rejection thresholds
recomputeNullRejectionThresholds();
//Restore the actual state of the null rejection
useNullRejection = tempUseNullRejection;
}else{
//Resize the rejection thresholds but set the values to 0
rejectionThresholds.clear();
rejectionThresholds.resize( numClasses, 0 );
}
return true;
}
void FilterBase::processMeasurement(const Measurement &measurement)
{
FB_DEBUG("------ FilterBase::processMeasurement (" << measurement.topicName_ << ") ------\n");
double delta = 0.0;
// If we've had a previous reading, then go through the predict/update
// cycle. Otherwise, set our state and covariance to whatever we get
// from this measurement.
if (initialized_)
{
// Determine how much time has passed since our last measurement
delta = measurement.time_ - lastMeasurementTime_;
FB_DEBUG("Filter is already initialized. Carrying out predict/correct loop...\n"
"Measurement time is " << std::setprecision(20) << measurement.time_ <<
", last measurement time is " << lastMeasurementTime_ << ", delta is " << delta << "\n");
// Only want to carry out a prediction if it's
// forward in time. Otherwise, just correct.
if (delta > 0)
{
validateDelta(delta);
predict(delta);
// Return this to the user
predictedState_ = state_;
}
correct(measurement);
}
else
{
FB_DEBUG("First measurement. Initializing filter.\n");
// Initialize the filter, but only with the values we're using
size_t measurementLength = measurement.updateVector_.size();
for (size_t i = 0; i < measurementLength; ++i)
{
state_[i] = (measurement.updateVector_[i] ? measurement.measurement_[i] : state_[i]);
}
// Same for covariance
for (size_t i = 0; i < measurementLength; ++i)
{
for (size_t j = 0; j < measurementLength; ++j)
{
estimateErrorCovariance_(i, j) = (measurement.updateVector_[i] && measurement.updateVector_[j] ?
measurement.covariance_(i, j) :
estimateErrorCovariance_(i, j));
}
}
initialized_ = true;
}
if (delta >= 0.0)
{
// Update the last measurement and update time.
// The measurement time is based on the time stamp of the
// measurement, whereas the update time is based on this
// node's current ROS time. The update time is used to
// determine if we have a sensor timeout, whereas the
// measurement time is used to calculate time deltas for
// prediction and correction.
lastMeasurementTime_ = measurement.time_;
}
FB_DEBUG("------ /FilterBase::processMeasurement (" << measurement.topicName_ << ") ------\n");
}
double RobotTracker::stuck(double time)
{
Matrix x = predict(time);
return bound(x.e(6,0), 0, 1);
}
float predict2(float*values, int num_values, bool returnDFval)
{
return predict(values, num_values, returnDFval);
}
void CameraRig::predictAndSetRotation(float time) {
glm::quat headRotation = predict(time);
setRotation(headRotation);
}
//返回预测的方向
double RobotTracker::direction(double time)
{
Matrix x = predict(time);
return x.e(2,0);
}
int main(int argc, char* argv[]){
//random time for random values later for weights and biases
srand(time( NULL));
//load the input files
FILE* trainingData = fopen(argv[1], "r");
if(argc < 4){
printf("Not enough arguments\n");
return 1;
}
if(trainingData == NULL){
printf("File not found!");
return 1;
}
int INPUTLAYERNODES = atoi(argv[2]);
int HIDDENLAYERNODES = atoi(argv[3]);
int OUTPUTLAYERNODES = atoi(argv[4]);
DATASET trainData[DATATOREAD];
char line[127];
int lineCount = 0;
while (fgets ( line, sizeof line, trainingData )!= NULL && lineCount < DATATOREAD){
int currentValueCount = 0;
char currentValue[10];
for(int i = 0; i < 10; i++){ currentValue[i] ==(char) 0;}
int currentValuePos = 0;
for(int i = 0; i < sizeof(line); i++){
if(line[i] != ',' && line[i] != '\n'){
currentValue[currentValuePos] = line[i];
currentValuePos++;
}else{
trainData[lineCount].value[currentValueCount] = atof(currentValue);
for(int j = 0; j < 10; j++){
currentValue[j] = (char) 0;
}
currentValueCount++;
currentValuePos = 0;
}
}
lineCount++;
}
normalize(trainData);
for(int i = 0; i < DATATOREAD; i++){
trainData[i].value[7] -= 1;
}
for(int i = 0; i < DATATOREAD; i++){
printf("Data %i | ", i);
for(int j = 0; j < INPUTLAYERNODES+1; j++){
printf("%f|",trainData[i].value[j]);
}
printf("\n");
}
//setting up the network
float weightsFromInputToHidden[INPUTLAYERNODES * HIDDENLAYERNODES];
for (int i = 0; i < INPUTLAYERNODES * HIDDENLAYERNODES; i++) {
weightsFromInputToHidden[i] = (float) rand() / RAND_MAX * 2.0 - 1.0;
}
float weightsFromHiddenToOutput[HIDDENLAYERNODES * OUTPUTLAYERNODES];
for (int i = 0; i < HIDDENLAYERNODES * OUTPUTLAYERNODES; i++) {
weightsFromHiddenToOutput[i] = (float) rand() / RAND_MAX * 2.0 - 1.0;
}
float biasesOfHidden[HIDDENLAYERNODES];
for (int i = 0; i < HIDDENLAYERNODES; i++) {
biasesOfHidden[i] = (float) rand() / RAND_MAX * 2.0 - 1.0;
}
float biasesOfOutput[OUTPUTLAYERNODES];
for (int i = 0; i < OUTPUTLAYERNODES; i++) {
biasesOfOutput[i] = (float) rand() / RAND_MAX * 2.0 - 1.0;
}
trainNetwork(trainData, weightsFromInputToHidden, weightsFromHiddenToOutput, biasesOfHidden, biasesOfOutput, INPUTLAYERNODES, HIDDENLAYERNODES, OUTPUTLAYERNODES);
fclose(trainingData);
printf("predicting...\n");
int right = 0;
for(int i = 0; i < DATATOREAD; i++){
float data[INPUTLAYERNODES+1];
data[0] = trainData[i].value[0];
data[1] = trainData[i].value[1];
data[2] = trainData[i].value[2];
data[3] = trainData[i].value[3];
data[4] = trainData[i].value[4];
data[5] = trainData[i].value[5];
data[6] = trainData[i].value[6];
data[7] = trainData[i].value[7];
right += predict(data, weightsFromInputToHidden, weightsFromHiddenToOutput, biasesOfHidden, biasesOfOutput, INPUTLAYERNODES, HIDDENLAYERNODES, OUTPUTLAYERNODES);
}
printf("I got %i / %i\n", right, DATATOREAD);
printf("end.\n");
return 0;
}
static av_always_inline void decode_line(FFV1Context *s, int w,
int16_t *sample[2],
int plane_index, int bits)
{
PlaneContext *const p = &s->plane[plane_index];
RangeCoder *const c = &s->c;
int x;
int run_count = 0;
int run_mode = 0;
int run_index = s->run_index;
if (s->slice_coding_mode == 1) {
int i;
for (x = 0; x < w; x++) {
int v = 0;
for (i=0; i<bits; i++) {
uint8_t state = 128;
v += v + get_rac(c, &state);
}
sample[1][x] = v;
}
return;
}
for (x = 0; x < w; x++) {
int diff, context, sign;
context = get_context(p, sample[1] + x, sample[0] + x, sample[1] + x);
if (context < 0) {
context = -context;
sign = 1;
} else
sign = 0;
av_assert2(context < p->context_count);
if (s->ac) {
diff = get_symbol_inline(c, p->state[context], 1);
} else {
if (context == 0 && run_mode == 0)
run_mode = 1;
if (run_mode) {
if (run_count == 0 && run_mode == 1) {
if (get_bits1(&s->gb)) {
run_count = 1 << ff_log2_run[run_index];
if (x + run_count <= w)
run_index++;
} else {
if (ff_log2_run[run_index])
run_count = get_bits(&s->gb, ff_log2_run[run_index]);
else
run_count = 0;
if (run_index)
run_index--;
run_mode = 2;
}
}
run_count--;
if (run_count < 0) {
run_mode = 0;
run_count = 0;
diff = get_vlc_symbol(&s->gb, &p->vlc_state[context],
bits);
if (diff >= 0)
diff++;
} else
diff = 0;
} else
diff = get_vlc_symbol(&s->gb, &p->vlc_state[context], bits);
av_dlog(s->avctx, "count:%d index:%d, mode:%d, x:%d pos:%d\n",
run_count, run_index, run_mode, x, get_bits_count(&s->gb));
}
if (sign)
diff = -diff;
sample[1][x] = (predict(sample[1] + x, sample[0] + x) + diff) &
((1 << bits) - 1);
}
s->run_index = run_index;
}
int main()
{
int i=0;
init_devices();
lcd_set_4bit();
lcd_init();
color_sensor_scaling();
/*
//variable 'i' scales at 13,14 for sharp sensor for velocitty 240 240
//u turn 1600ms at 200,200 velocity
velocity(200,200);
left();
_delay_ms(1600);
stop();
while(1);
threshold=6000;
right();
while(ADC_Conversion(11)<65)
{
i++;
lcd_print(1,11,i,3);
}
stop();
lcd_print(2,11,scan(),1);
stop();
while(1);
*/
setIndicatorAndColor();
threshold=6000;
ct = 0; adj = 2;
//lcd("Begin");
forwardJaa();
stop();
servo_1(0);
servo_2(90);
servo_3(0);
while (sorted<total)
{
canDrop();
//buzzer_on();
//_delay_ms(500);
//buzzer_off();
if (visitedCount == 3)
predict();
if (sorted == total)
break;
pickup();
traverseToSort(ct, ct % 2 + 4);
sortCheck();
}
for (i = 0; i<4; i++);
while(1);
//..printf("%d %d\n", term[i][0], term[i][1]);
//..printf("Sort 0=%dSort 1=%d\nArm 0=%dArm 1=%d\n", sort[0], sort[1], arm[0], arm[1]);
//..printf("Cost=%d\nSORTED!!!!!\n", cost + 7);
//getch();
return 0;
}
void binary_class_predict(FILE *input, FILE *output){
int total = 0;
int *labels;
int max_nr_attr = 64;
struct feature_node *x = Malloc(struct feature_node, max_nr_attr);
dvec_t dec_values;
ivec_t true_labels;
int n;
if(model_->bias >= 1)
n = get_nr_feature(model_) + 1;
else
n = get_nr_feature(model_);
labels = Malloc(int, get_nr_class(model_));
get_labels(model_, labels);
max_line_len = 1024;
line = (char *)malloc(max_line_len*sizeof(char));
while(readline(input) != NULL)
{
int i = 0;
double target_label, predict_label;
char *idx, *val, *label, *endptr;
int inst_max_index = -1; // strtol gives 0 if wrong format, and precomputed kernel has <index> start from 0
label = strtok(line," \t");
target_label = strtod(label,&endptr);
if(endptr == label)
exit_input_error(total+1);
while(1)
{
if(i>=max_nr_attr - 2) // need one more for index = -1
{
max_nr_attr *= 2;
x = (struct feature_node *) realloc(x,max_nr_attr*sizeof(struct feature_node));
}
idx = strtok(NULL,":");
val = strtok(NULL," \t");
if(val == NULL)
break;
errno = 0;
x[i].index = (int) strtol(idx,&endptr,10);
if(endptr == idx || errno != 0 || *endptr != '\0' || x[i].index <= inst_max_index)
exit_input_error(total+1);
else
inst_max_index = x[i].index;
errno = 0;
x[i].value = strtod(val,&endptr);
if(endptr == val || errno != 0 || (*endptr != '\0' && !isspace(*endptr)))
exit_input_error(total+1);
++i;
}
if(model_->bias >= 0){
x[i].index = n;
x[i].value = model_->bias;
++i;
}
x[i].index = -1;
predict_label = predict(model_,x);
fprintf(output,"%g\n",predict_label);
double dec_value;
predict_values(model_, x, &dec_value);
true_labels.push_back((target_label > 0)? 1: -1);
if(labels[0] <= 0) dec_value *= -1;
dec_values.push_back(dec_value);
}
validation_function(dec_values, true_labels);
free(labels);
free(x);
}
void do_predict(FILE *input, FILE *output)
{
int total=0;
int n;
int nr_feature=get_nr_feature(model_);
double *dvec_t;
double *ivec_t;
int *query;
n=nr_feature;
max_line_len = 1024;
line = (char *)malloc(max_line_len*sizeof(char));
while(readline(input) != NULL)
total++;
rewind(input);
dvec_t = new double[total];
ivec_t = new double[total];
query = new int[total];
total = 0;
while(readline(input) != NULL)
{
int i = 0;
double target_label, predict_label;
char *idx, *val, *label, *endptr;
int inst_max_index = 0; // strtol gives 0 if wrong format
query[total] = 0;
label = strtok(line," \t\n");
if(label == NULL) // empty line
exit_input_error(total+1);
target_label = strtod(label,&endptr);
if(endptr == label || *endptr != '\0')
exit_input_error(total+1);
ivec_t[total] = target_label;
while(1)
{
if(i>=max_nr_attr-2) // need one more for index = -1
{
max_nr_attr *= 2;
x = (struct feature_node *) realloc(x,max_nr_attr*sizeof(struct feature_node));
}
idx = strtok(NULL,":");
val = strtok(NULL," \t");
if(val == NULL)
break;
if (strcmp(idx,"qid") == 0)
{
errno = 0;
query[total] = (int) strtol(val,&endptr,10);
if(endptr == val || errno != 0 || *endptr != '\0')
exit_input_error(i+1);
continue;
}
errno = 0;
x[i].index = (int) strtol(idx,&endptr,10);
if(endptr == idx || errno != 0 || *endptr != '\0' || x[i].index <= inst_max_index)
exit_input_error(total+1);
else
inst_max_index = x[i].index;
errno = 0;
x[i].value = strtod(val,&endptr);
if(endptr == val || errno != 0 || (*endptr != '\0' && !isspace(*endptr)))
exit_input_error(total+1);
// feature indices larger than those in training are not used
if(x[i].index <= nr_feature)
++i;
}
x[i].index = -1;
predict_label = predict(model_,x);
fprintf(output,"%.10f\n",predict_label);
dvec_t[total++] = predict_label;
}
double result[3];
eval_list(ivec_t,dvec_t,query,total,result);
info("Pairwise Accuracy = %g%%\n",result[0]*100);
info("MeanNDCG (LETOR) = %g\n",result[1]);
info("NDCG (YAHOO) = %g\n",result[2]);
}
// Optimise the two lagrange multipliers, if successful then return 1
int SMO::takeStep(int i1, int i2)
{
// Old values of alpha[i1] and alpha[i2]
double alpha1, alpha2;
// New values for alpha[i1] and alpha[i2]
double a1, a2;
int y1, y2, s;
double E1, E2, L, H, k11, k22, k12, eta, Lobj, Hobj, dT;
if (i1 == i2) return 0;
alpha1 = alpha[i1];
y1 = target[i1];
// Check E1 in error cache, if not present then E1 = SVM output on point[i1] - y1.
if (alpha1 > 0 && alpha1 < C)
E1 = errorCache[i1];
else
E1 = predict(points[i1]) - y1;
alpha2 = alpha[i2];
y2 = target[i2];
if (alpha2 > 0 && alpha2 < C)
E2 = errorCache[i2];
else
E2 = predict(points[i2]) - y2;
s = y1 * y2;
// Compute L and H
if (y1 == y2)
{
L = std::max(0.0, alpha1 + alpha2 - C);
H = std::min(C, alpha1 + alpha2);
}
else
{
L = std::max(0.0, alpha2 - alpha1);
H = std::min(C, alpha2 - alpha1 + C);
}
if (L == H)
return 0;
k11 = kernel(points[i1], points[i1]);
k12 = kernel(points[i1], points[i2]);
k22 = kernel(points[i2], points[i2]);
eta = 2 * k12 - k11 - k22;
if (eta < 0)
{
a2 = alpha2 + y2 * (E2 - E1) / eta;
if (a2 < L) a2 = L;
else if (a2 > H) a2 = H;
}
else
{
double c1 = eta/2;
double c2 = y2 * (E1-E2)- eta * alpha2;
Lobj = c1 * L * L + c2 * L;
Hobj = c1 * H * H + c2 * H;
if (Lobj > Hobj+epsilon) a2 = L;
else if (Lobj < Hobj-epsilon) a2 = H;
else a2 = alpha2;
}
if (fabs(a2-alpha2) < epsilon*(a2+alpha2+epsilon))
return 0;
a1 = alpha1 - s * (a2 - alpha2);
if (a1 < 0)
{
a2 += s * a1;
a1 = 0;
}
else if (a1 > C)
{
double t = a1-C;
a2 += s * t;
a1 = C;
}
double b1, b2, newThreshold;
if (a1 > 0 && a1 < C)
newThreshold = threshold + E1 + y1 * (a1 - alpha1) * k11 + y2 * (a2 - alpha2) * k12;
else
{
if (a2 > 0 && a2 < C)
newThreshold = threshold + E2 + y1 * (a1 - alpha1) * k12 + y2 * (a2 - alpha2) * k22;
else
{
b1 = threshold + E1 + y1 * (a1 - alpha1) * k11 + y2 * (a2 - alpha2) * k12;
b2 = threshold + E2 + y1 * (a1 - alpha1) * k12 + y2 * (a2 - alpha2) * k22;
newThreshold = (b1 + b2) / 2;
}
}
// Change in threshold value
dT = newThreshold - threshold;
// Update threshold to reflect changes in lagrange multipliers.
threshold = newThreshold;
double t1 = y1 * (a1-alpha1);
double t2 = y2 * (a2-alpha2);
// For linear kernel update weights to reflect changes in a1 and a2.
for (unsigned int i=0; i < points[0].size(); i++)
w[i] += points[i1][i] * t1 + points[i2][i] * t2;
// Update error cache using new lagrange's multipliers.
for (unsigned int i=0; i<points.size(); i++)
if (0 < alpha[i] && alpha[i] < C)
errorCache[i] += t1 * kernel(points[i1],points[i]) + t2 * kernel(points[i2],points[i]) - dT;
errorCache[i1] = 0.0;
errorCache[i2] = 0.0;
// Update alpha with a1 and a2
alpha[i1] = a1;
alpha[i2] = a2;
return 1;
}
int SMO::examineExample(int i1)
{
double y1, alpha1, E1, r1;
y1 = target[i1];
alpha1 = alpha[i1];
// Check E1 in error cache, if not present then E1 = SVM output on point[i1] - y1.
if (alpha1 > 0 && alpha1 < C)
E1 = errorCache[i1];
else
E1 = predict(points[i1]) - y1;
r1 = y1 * E1;
// Check if alpha1 violated KKT condition by more than tolerance, if it does then look for alpha2 and optimise them by calling takeStep(i1,i2)
if ((r1 < -tolerance && alpha1 < C)
|| (r1 > tolerance && alpha1 > 0))
{
unsigned int k;
int i2;
double dEmax = 0.0;
i2 = -1;
// Try i2 using second choice heuristic as described in section 2.2 by choosing an error to maximize step size.
// The i2 is taken which maximize dE.
for (k = 0; k < points.size(); k++)
if (alpha[k] > 0 && alpha[k] < C)
{
double E2, dE;
E2 = errorCache[k];
dE = fabs(E1 - E2);
if (dE > dEmax)
{
dEmax = dE;
i2 = k;
}
}
if (i2 >= 0)
{
if (takeStep (i1, i2))
return 1;
}
// Loop over all non-zero and non-C alpha, starting at a random point.
i2 = (int)(rand() % points.size());
for (k = 0; k < points.size(); k++)
{
if (alpha[i2] > 0 && alpha[i2] < C)
{
if (takeStep(i1, i2))
return 1;
}
i2 = (i2 + 1)%points.size();
}
// Loop over all possible i2, starting at a random point.
i2 = (int)(rand() % points.size());
for (k = 0; k < points.size(); k++)
{
if (takeStep(i1, i2))
return 1;
i2 = (i2 + 1)%points.size();
}
}
return 0;
}
void Segmentor::train(const string& trainFile, const string& devFile, const string& testFile, const string& modelFile, const string& optionFile,
const string& wordEmbFile) {
if (optionFile != "")
m_options.load(optionFile);
m_options.showOptions();
vector<Instance> trainInsts, devInsts, testInsts;
m_pipe.readInstances(trainFile, trainInsts, m_classifier.MAX_SENTENCE_SIZE - 2, m_options.maxInstance);
if (devFile != "")
m_pipe.readInstances(devFile, devInsts, m_classifier.MAX_SENTENCE_SIZE - 2, m_options.maxInstance);
if (testFile != "")
m_pipe.readInstances(testFile, testInsts, m_classifier.MAX_SENTENCE_SIZE - 2, m_options.maxInstance);
vector<vector<Instance> > otherInsts(m_options.testFiles.size());
for (int idx = 0; idx < m_options.testFiles.size(); idx++) {
m_pipe.readInstances(m_options.testFiles[idx], otherInsts[idx], m_classifier.MAX_SENTENCE_SIZE - 2, m_options.maxInstance);
}
createAlphabet(trainInsts);
addTestWordAlpha(devInsts);
addTestWordAlpha(testInsts);
for (int idx = 0; idx < otherInsts.size(); idx++) {
addTestWordAlpha(otherInsts[idx]);
}
m_classifier.init();
m_classifier.setDropValue(m_options.dropProb);
vector<vector<CAction> > trainInstGoldactions;
getGoldActions(trainInsts, trainInstGoldactions);
double bestFmeasure = 0;
int inputSize = trainInsts.size();
std::vector<int> indexes;
for (int i = 0; i < inputSize; ++i)
indexes.push_back(i);
static Metric eval, metric_dev, metric_test;
int maxIter = m_options.maxIter * (inputSize / m_options.batchSize + 1);
int oneIterMaxRound = (inputSize + m_options.batchSize -1) / m_options.batchSize;
std::cout << "maxIter = " << maxIter << std::endl;
int devNum = devInsts.size(), testNum = testInsts.size();
static vector<vector<string> > decodeInstResults;
static vector<string> curDecodeInst;
static bool bCurIterBetter;
static vector<vector<string> > subInstances;
static vector<vector<CAction> > subInstGoldActions;
for (int iter = 0; iter < maxIter; ++iter) {
std::cout << "##### Iteration " << iter << std::endl;
srand(iter);
random_shuffle(indexes.begin(), indexes.end());
std::cout << "random: " << indexes[0] << ", " << indexes[indexes.size() - 1] << std::endl;
bool bEvaluate = false;
if(m_options.batchSize == 1){
eval.reset();
bEvaluate = true;
for (int idy = 0; idy < inputSize; idy++) {
subInstances.clear();
subInstGoldActions.clear();
subInstances.push_back(trainInsts[indexes[idy]].chars);
subInstGoldActions.push_back(trainInstGoldactions[indexes[idy]]);
double cost = m_classifier.train(subInstances, subInstGoldActions);
eval.overall_label_count += m_classifier._eval.overall_label_count;
eval.correct_label_count += m_classifier._eval.correct_label_count;
if ((idy + 1) % (m_options.verboseIter*10) == 0) {
std::cout << "current: " << idy + 1 << ", Cost = " << cost << ", Correct(%) = " << eval.getAccuracy() << std::endl;
}
m_classifier.updateParams(m_options.regParameter, m_options.adaAlpha, m_options.adaEps);
}
std::cout << "current: " << iter + 1 << ", Correct(%) = " << eval.getAccuracy() << std::endl;
}
else{
if(iter == 0)eval.reset();
subInstances.clear();
subInstGoldActions.clear();
for (int idy = 0; idy < m_options.batchSize; idy++) {
subInstances.push_back(trainInsts[indexes[idy]].chars);
subInstGoldActions.push_back(trainInstGoldactions[indexes[idy]]);
}
double cost = m_classifier.train(subInstances, subInstGoldActions);
eval.overall_label_count += m_classifier._eval.overall_label_count;
eval.correct_label_count += m_classifier._eval.correct_label_count;
if ((iter + 1) % (m_options.verboseIter) == 0) {
std::cout << "current: " << iter + 1 << ", Cost = " << cost << ", Correct(%) = " << eval.getAccuracy() << std::endl;
eval.reset();
bEvaluate = true;
}
m_classifier.updateParams(m_options.regParameter, m_options.adaAlpha, m_options.adaEps);
}
if (bEvaluate && devNum > 0) {
bCurIterBetter = false;
if (!m_options.outBest.empty())
decodeInstResults.clear();
metric_dev.reset();
for (int idx = 0; idx < devInsts.size(); idx++) {
predict(devInsts[idx], curDecodeInst);
devInsts[idx].evaluate(curDecodeInst, metric_dev);
if (!m_options.outBest.empty()) {
decodeInstResults.push_back(curDecodeInst);
}
}
std::cout << "dev:" << std::endl;
metric_dev.print();
if (!m_options.outBest.empty() && metric_dev.getAccuracy() > bestFmeasure) {
m_pipe.outputAllInstances(devFile + m_options.outBest, decodeInstResults);
bCurIterBetter = true;
}
if (testNum > 0) {
if (!m_options.outBest.empty())
decodeInstResults.clear();
metric_test.reset();
for (int idx = 0; idx < testInsts.size(); idx++) {
predict(testInsts[idx], curDecodeInst);
testInsts[idx].evaluate(curDecodeInst, metric_test);
if (bCurIterBetter && !m_options.outBest.empty()) {
decodeInstResults.push_back(curDecodeInst);
}
}
std::cout << "test:" << std::endl;
metric_test.print();
if (!m_options.outBest.empty() && bCurIterBetter) {
m_pipe.outputAllInstances(testFile + m_options.outBest, decodeInstResults);
}
}
for (int idx = 0; idx < otherInsts.size(); idx++) {
std::cout << "processing " << m_options.testFiles[idx] << std::endl;
if (!m_options.outBest.empty())
decodeInstResults.clear();
metric_test.reset();
for (int idy = 0; idy < otherInsts[idx].size(); idy++) {
predict(otherInsts[idx][idy], curDecodeInst);
otherInsts[idx][idy].evaluate(curDecodeInst, metric_test);
if (bCurIterBetter && !m_options.outBest.empty()) {
decodeInstResults.push_back(curDecodeInst);
}
}
std::cout << "test:" << std::endl;
metric_test.print();
if (!m_options.outBest.empty() && bCurIterBetter) {
m_pipe.outputAllInstances(m_options.testFiles[idx] + m_options.outBest, decodeInstResults);
}
}
if (m_options.saveIntermediate && metric_dev.getAccuracy() > bestFmeasure) {
std::cout << "Exceeds best previous DIS of " << bestFmeasure << ". Saving model file.." << std::endl;
bestFmeasure = metric_dev.getAccuracy();
writeModelFile(modelFile);
}
}
}
}
int main(int argc, char* argv[])
{
// Read parameters from console.
CCmdLine cmdLine;
if(cmdLine.SplitLine(argc, argv) < 5)
{
cerr << "Usage: ./bbnet -s score_file -n node -b bkg -f func_depth -o output" << endl;
cerr << endl << "Additional parameter:" << endl;
cerr << "-k\tPenalty parameter(logK, Default = 5.0)" << endl;
cerr << "-c\tnumber of candidate motifs (Default=50)" << endl;
cerr << "-d\tpositive negative (for prediction using BN)" << endl;
cerr << "-l\toutput of all training samples' information." << endl;
cerr << "-t\ttranslational(transcriptional) start sites.(Default = right end)" << endl;
cerr << "-rb\tbit-string to determine which rules to include.(Default = 111110)" << endl;
cerr << "-i\tUse mutual information instead of Bayesian score" << endl;
cerr << endl << "Contact: \"Li Shen\"<*****@*****.**>" << endl;
return 1;
}
string s, n, b, f, o;
try
{
s = cmdLine.GetArgument("-s", 0); // score file.
n = cmdLine.GetArgument("-n", 0); // node gene list file.
b = cmdLine.GetArgument("-b", 0); // bkg gene list file.
f = cmdLine.GetArgument("-f", 0); // func depth folder.
o = cmdLine.GetArgument("-o", 0); // results output file.
}
catch(int)
{
cerr << "Wrong arguments!" << endl;
return 1;
}
if(cmdLine.HasSwitch("-i"))
itag = true;
//itag = true;
//string s = "../gbnet/data/Beer/scor_test.list";
//string n = "../gbnet/data/Beer/node.list";
//string b = "../gbnet/data/Beer/bkg.list";
//string f = "../gbnet/data/Beer/func";
//string k = "0.015";
//logK = atof(k.data());
//string o = "../gbnet/data/Beer/bb_res_test2.txt";
string k; // Penalty parameter; logK value.
if(!itag)
k = cmdLine.GetSafeArgument("-k", 0, "5.0");
else
k = cmdLine.GetSafeArgument("-k", 0, "0.015");
logK = atof(k.data());
string c = cmdLine.GetSafeArgument("-c", 0, "50"); // number of candidate motifs. default = 50.
motifcand = atoi(c.data());
// Use prior counts for some motifs if specified.
string p = cmdLine.GetSafeArgument("-p", 0, "0"); // prior counts.
pricnt = atoi(p.data());
if(pricnt > 0) // read preferred motifs list from file.
{
prior = 1;
string fPrim = cmdLine.GetSafeArgument("-p", 1, "primot.txt");
vector<string> primv;
if(get1stcol(fPrim, primv) < 0)
return 1;
for(size_t i = 0; i < primv.size(); i++)
primo.insert(primv[i]);
}
// File names for positive, negative and left-out testing lists.
string pos = cmdLine.GetSafeArgument("-d", 0, ""); // positive testing cases.
string neg = cmdLine.GetSafeArgument("-d", 1, ""); // negative testing cases.
string res = cmdLine.GetSafeArgument("-d", 2, ""); // left-out testing cases.
vector<string> plst, nlst, rlst; // positive, negative and left-out lists.
if(pos != "" && neg != "")
{
if(get1stcol(pos, plst) < 0)
return 1;
if(get1stcol(neg, nlst) < 0)
return 1;
}
if(res != "")
{
if(get1stcol(res, rlst) < 0)
return 1;
}
// File for output of all training samples' information.
string finfo = cmdLine.GetSafeArgument("-l", 0, "");
// File to store all genes' translational/transcriptional start sites.
string ftss = cmdLine.GetSafeArgument("-t", 0, "");
if(ftss != "")
loadtss(ftss, mtss);
// A bit-string to determine which rules to include.
rb = cmdLine.GetSafeArgument("-rb", 0, "111110");
string bp = cmdLine.GetSafeArgument("-bp", 0, ""); // Output each gene's probability like in Beer's prediction.
// Load motif Bayesian score file.
if(loadscor(mscor, s) != 0)
{
cerr << "Load motif scores eror!" << endl;
return 1;
}
else
{
#ifdef VERBOSE
cout << "Display candidate motifs that are loaded:" << endl;
dispscor(mscor);
#endif
}
vector<MotifScore> oscor = mscor; // Save an original copy of motif scores.
// Load gene list.
vector<Case> tlst, blst, genlst;
set<string> genset;
if(loadgene(tlst, blst, n, b) != 0)
{
cerr << "Load gene lists error!" << endl;
return 1;
}
else
{
genlst.insert(genlst.end(), tlst.begin(), tlst.end());
genlst.insert(genlst.end(), blst.begin(), blst.end());
#ifdef VERBOSE
cout << "Load gene list completed!" << endl;
#endif
// All training and testing gene names are put into genmap.
for(size_t i = 0; i < genlst.size(); i++)
genset.insert(genlst[i].name);
for(size_t i = 0; i < plst.size(); i++)
genset.insert(plst[i]);
for(size_t i = 0; i < nlst.size(); i++)
genset.insert(nlst[i]);
for(size_t i = 0; i < rlst.size(); i++)
genset.insert(rlst[i]);
}
// Load motif binding information of genes in genmap.
if(loadbind(allbind, mscor, genset, f) != 0)
{
cerr << "Load binding information error!" << endl;
return 1;
}
else
{
#ifdef VERBOSE
cout << "Load binding information completed!" << endl;
#endif
}
// File for output.
ofstream hOut(o.data());
if(!hOut)
{
cerr << "Can't open " << o << endl;
return 1;
}
hOut << "Number of genes in category 1: " << tlst.size() << endl;
hOut << "Number of genes in category 0: " << blst.size() << endl << endl;
#ifdef VERBOSE
cout << endl << "Running on original data." << endl;
#endif
vector<Constraint> cons;
vector<CPTRow> cpt;
clock_t start = clock();
double scor = bbnet(cons, cpt, genlst);
clock_t finish = clock();
if(outbayes(hOut, scor, cons, cpt, oscor, tlst.size(), blst.size()) != 0)
{
cerr << "Output Bayesian network results error!" << endl;
return 1;
}
if(finfo != "")
{
if(outgene(finfo, tlst, blst, cons) != 0)
cerr << "Output training samples' information error!" << endl;
return 1;
}
if(pos != "" && neg != "")
{
Pred d = predict(cons, cpt, plst, nlst);
outpred(hOut, d, n, b, pos, neg);
if(bp != "") // output each gene's probability being in this cluster if output file is specified.
{
ofstream hbp(bp.data());
if(!hbp)
{
cerr << "Can't open " << bp << endl;
return 1;
}
vector<BPred> trnbp = predict(cons, cpt, genlst, 0); // probabilities for training genes.
outpred(hbp, trnbp);
vector<string> tstlst; // probabilities for testing genes.
tstlst.insert(tstlst.end(), plst.begin(), plst.end()); // positive testings.
tstlst.insert(tstlst.end(), nlst.begin(), nlst.end()); // negative testings.
vector<BPred> tstbp = predict(cons, cpt, tstlst, 1);
outpred(hbp, tstbp);
if(res != "") // probabilities for left-out genes if the left-out file is specified.
{
vector<BPred> lefbp = predict(cons, cpt, rlst, -1);
outpred(hbp, lefbp);
}
hbp.close();
}
}
hOut << endl << "Bayesian network occupied CPU " << (double)(finish-start)/CLOCKS_PER_SEC << " seconds." << endl;
hOut.close();
return 0;
}
void MainWindow::createActions() {
_openiTunesAction = new QAction(QIcon(":/images/iTunes.png"),
tr("Open i&Tunes Library..."), this);
_openiTunesAction->setShortcut(tr("Ctrl+T"));
_openiTunesAction->setStatusTip(tr("Open iTunes Library file"));
connect(_openiTunesAction, SIGNAL(triggered()), this, SLOT(openiTunesLibrary()));
_openCollectionAction = new QAction(QIcon(":/images/openCollection.png"),
tr("Open Collection File"), this);
_openCollectionAction->setShortcut(tr("Ctrl+T"));
_openCollectionAction->setStatusTip(tr("Open Collection file"));
connect(_openCollectionAction, SIGNAL(triggered()), this, SLOT(openCollectionFile()));
_exitAction = new QAction(tr("E&xit"), this);
_exitAction->setShortcut(tr("Ctrl+Q"));
connect(_exitAction, SIGNAL(triggered()), this, SLOT(close()));
_coutAction = new QAction(tr("&Cout Library"), this);
_coutAction->setShortcut(tr("Ctrl+E"));
connect(_coutAction, SIGNAL(triggered()), this, SLOT(display()));
_predictAction = new QAction(QIcon(":/images/predict.png"),
tr("&Predict"), this);
_predictAction->setShortcut(tr("Ctrl+3"));
_predictAction->setStatusTip(tr("Predict the the placement of the prediction tracks"));
connect(_predictAction, SIGNAL(triggered()), _display, SLOT(predict()));
_extractAction = new QAction(QIcon(":/images/extract.png"),
tr("&Extract"), this);
_extractAction->setShortcut(tr("Ctrl+1"));
_extractAction->setStatusTip(tr("Extract features from the defined training tracks"));
connect(_extractAction, SIGNAL(triggered()), _display, SLOT(extract()));
_trainingAction = new QAction(QIcon(":/images/train.png"),
tr("&Train"), this);
_trainingAction->setShortcut(tr("Ctrl+2"));
_trainingAction->setStatusTip(tr("Train the grid using the defined training tracks"));
connect(_trainingAction, SIGNAL(triggered()), _display, SLOT(train()));
_initAction = new QAction(QIcon(":/images/init.png"), tr("&Initlize"), this);
connect(_initAction, SIGNAL(triggered()), _display, SLOT(init()));
_aboutAction = new QAction(tr("&About"), this);
connect(_aboutAction, SIGNAL(triggered()), this, SLOT(about()));
_saveGridAction = new QAction(tr("&Save Grid"), this);
connect(_saveGridAction, SIGNAL(triggered()), this, SLOT(saveCurrentGrid()) );
_loadGridAction = new QAction(tr("&Load Saved Grid"),this);
connect(_loadGridAction, SIGNAL(triggered()), this, SLOT(openSavedGrid()));
_playModeAction = new QAction(tr("&Continuous"), this);
connect(_playModeAction, SIGNAL(triggered()), this, SLOT(changedPlayMode()));
_cancelAction = new QAction(tr("&Cancel Action"), this);
connect(_cancelAction, SIGNAL(triggered()), this, SLOT(cancelButton()));
_normHashAction = new QAction(tr("Open Saved Hash"), this);
connect(_normHashAction, SIGNAL(triggered()), _display, SLOT(hashLoad()));
_fullScreenAction = new QAction (tr("&Full Screen Mouse Mode"), this);
connect(_fullScreenAction, SIGNAL(triggered()), _display, SLOT(fullScreenMouse()));
_colourMapMode = new QAction (tr("&Colour Mapping Mode"),this);
connect(_colourMapMode, SIGNAL(triggered()), _display, SLOT(colourMapMode()));
_resetButtonAction = new QAction (tr("&Reset"), this);
connect (_resetButtonAction, SIGNAL(triggered()), this, SLOT(resetButtonPressed()));
_optionsDialogAction = new QAction(tr("&Options"), this);
connect(_optionsDialogAction, SIGNAL(triggered()), this, SLOT(optionsDialogTriggered()));
}
bool KNN::train_(const ClassificationData &trainingData,const UINT K){
//Set the dimensionality of the input data
this->K = K;
//Flag that the algorithm has been trained so we can compute the rejection thresholds
trained = true;
//If null rejection is enabled then compute the null rejection thresholds
if( useNullRejection ){
//Set the null rejection to false so we can compute the values for it (this will be set back to its current value later)
useNullRejection = false;
nullRejectionThresholds.clear();
//Compute the rejection thresholds for each of the K classes
VectorFloat counter(numClasses,0);
trainingMu.resize( numClasses, 0 );
trainingSigma.resize( numClasses, 0 );
nullRejectionThresholds.resize( numClasses, 0 );
//Compute Mu for each of the classes
const unsigned int numTrainingExamples = trainingData.getNumSamples();
Vector< IndexedDouble > predictionResults( numTrainingExamples );
for(UINT i=0; i<numTrainingExamples; i++){
predict( trainingData[i].getSample(), K);
UINT classLabelIndex = 0;
for(UINT k=0; k<numClasses; k++){
if( predictedClassLabel == classLabels[k] ){
classLabelIndex = k;
break;
}
}
predictionResults[ i ].index = classLabelIndex;
predictionResults[ i ].value = classDistances[ classLabelIndex ];
trainingMu[ classLabelIndex ] += predictionResults[ i ].value;
counter[ classLabelIndex ]++;
}
for(UINT j=0; j<numClasses; j++){
trainingMu[j] /= counter[j];
}
//Compute Sigma for each of the classes
for(UINT i=0; i<numTrainingExamples; i++){
trainingSigma[predictionResults[i].index] += SQR(predictionResults[i].value - trainingMu[predictionResults[i].index]);
}
for(UINT j=0; j<numClasses; j++){
Float count = counter[j];
if( count > 1 ){
trainingSigma[ j ] = sqrt( trainingSigma[j] / (count-1) );
}else{
trainingSigma[ j ] = 1.0;
}
}
//Check to see if any of the mu or sigma values are zero or NaN
bool errorFound = false;
for(UINT j=0; j<numClasses; j++){
if( trainingMu[j] == 0 ){
warningLog << "TrainingMu[ " << j << " ] is zero for a K value of " << K << std::endl;
}
if( trainingSigma[j] == 0 ){
warningLog << "TrainingSigma[ " << j << " ] is zero for a K value of " << K << std::endl;
}
if( grt_isnan( trainingMu[j] ) ){
errorLog << "TrainingMu[ " << j << " ] is NAN for a K value of " << K << std::endl;
errorFound = true;
}
if( grt_isnan( trainingSigma[j] ) ){
errorLog << "TrainingSigma[ " << j << " ] is NAN for a K value of " << K << std::endl;
errorFound = true;
}
}
if( errorFound ){
trained = false;
return false;
}
//Compute the rejection thresholds
for(unsigned int j=0; j<numClasses; j++){
nullRejectionThresholds[j] = trainingMu[j] + (trainingSigma[j]*nullRejectionCoeff);
}
//Restore the actual state of the null rejection
useNullRejection = true;
}else{
//Resize the rejection thresholds but set the values to 0
nullRejectionThresholds.clear();
nullRejectionThresholds.resize( numClasses, 0 );
}
return true;
}
void Labeler::train(const string& trainFile, const string& devFile, const string& testFile, const string& modelFile, const string& optionFile,
const string& wordEmbFile, const string& charEmbFile) {
if (optionFile != "")
m_options.load(optionFile);
m_options.showOptions();
m_linearfeat = 0;
vector<Instance> trainInsts, devInsts, testInsts;
static vector<Instance> decodeInstResults;
static Instance curDecodeInst;
bool bCurIterBetter = false;
m_pipe.readInstances(trainFile, trainInsts, m_options.maxInstance);
if (devFile != "")
m_pipe.readInstances(devFile, devInsts, m_options.maxInstance);
if (testFile != "")
m_pipe.readInstances(testFile, testInsts, m_options.maxInstance);
//Ensure that each file in m_options.testFiles exists!
vector<vector<Instance> > otherInsts(m_options.testFiles.size());
for (int idx = 0; idx < m_options.testFiles.size(); idx++) {
m_pipe.readInstances(m_options.testFiles[idx], otherInsts[idx], m_options.maxInstance);
}
//std::cout << "Training example number: " << trainInsts.size() << std::endl;
//std::cout << "Dev example number: " << trainInsts.size() << std::endl;
//std::cout << "Test example number: " << trainInsts.size() << std::endl;
createAlphabet(trainInsts);
if (!m_options.wordEmbFineTune) {
addTestWordAlpha(devInsts);
addTestWordAlpha(testInsts);
for (int idx = 0; idx < otherInsts.size(); idx++) {
addTestWordAlpha(otherInsts[idx]);
}
cout << "Remain words num: " << m_wordAlphabet.size() << endl;
}
if (!m_options.charEmbFineTune) {
addTestCharAlpha(devInsts);
addTestCharAlpha(testInsts);
for (int idx = 0; idx < otherInsts.size(); idx++) {
addTestCharAlpha(otherInsts[idx]);
}
cout << "Remain char num: " << m_charAlphabet.size() << endl;
}
NRMat<dtype> wordEmb;
if (wordEmbFile != "") {
readWordEmbeddings(wordEmbFile, wordEmb);
} else {
wordEmb.resize(m_wordAlphabet.size(), m_options.wordEmbSize);
wordEmb.randu(1000);
}
NRMat<dtype> charEmb;
if (charEmbFile != "") {
readWordEmbeddings(charEmbFile, charEmb);
} else {
charEmb.resize(m_charAlphabet.size(), m_options.charEmbSize);
charEmb.randu(1001);
}
m_classifier.init(wordEmb, charEmb, m_options.wordcontext, m_options.charcontext, m_labelAlphabet.size(), m_options.wordHiddenSize, m_options.charHiddenSize,
m_options.hiddenSize);
m_classifier.resetRemove(m_options.removePool);
m_classifier.setDropValue(m_options.dropProb);
m_classifier.setWordEmbFinetune(m_options.wordEmbFineTune);
vector<Example> trainExamples, devExamples, testExamples;
initialExamples(trainInsts, trainExamples);
initialExamples(devInsts, devExamples);
initialExamples(testInsts, testExamples);
vector<int> otherInstNums(otherInsts.size());
vector<vector<Example> > otherExamples(otherInsts.size());
for (int idx = 0; idx < otherInsts.size(); idx++) {
initialExamples(otherInsts[idx], otherExamples[idx]);
otherInstNums[idx] = otherExamples[idx].size();
}
dtype bestDIS = 0;
int inputSize = trainExamples.size();
srand(0);
std::vector<int> indexes;
for (int i = 0; i < inputSize; ++i)
indexes.push_back(i);
static Metric eval, metric_dev, metric_test;
static vector<Example> subExamples;
int devNum = devExamples.size(), testNum = testExamples.size();
int maxIter = m_options.maxIter;
if (m_options.batchSize > 1)
maxIter = m_options.maxIter * (inputSize / m_options.batchSize + 1);
dtype cost = 0.0;
std::cout << "maxIter = " << maxIter << std::endl;
for (int iter = 0; iter < m_options.maxIter; ++iter) {
std::cout << "##### Iteration " << iter << std::endl;
eval.reset();
if (m_options.batchSize == 1) {
random_shuffle(indexes.begin(), indexes.end());
for (int updateIter = 0; updateIter < inputSize; updateIter++) {
subExamples.clear();
int start_pos = updateIter;
int end_pos = (updateIter + 1);
if (end_pos > inputSize)
end_pos = inputSize;
for (int idy = start_pos; idy < end_pos; idy++) {
subExamples.push_back(trainExamples[indexes[idy]]);
}
int curUpdateIter = iter * inputSize + updateIter;
cost = m_classifier.process(subExamples, curUpdateIter);
eval.overall_label_count += m_classifier._eval.overall_label_count;
eval.correct_label_count += m_classifier._eval.correct_label_count;
if ((curUpdateIter + 1) % m_options.verboseIter == 0) {
//m_classifier.checkgrads(subExamples, curUpdateIter+1);
std::cout << "current: " << updateIter + 1 << ", total instances: " << inputSize << std::endl;
std::cout << "Cost = " << cost << ", SA Correct(%) = " << eval.getAccuracy() << std::endl;
}
m_classifier.updateParams(m_options.regParameter, m_options.adaAlpha, m_options.adaEps);
}
} else {
cost = 0.0;
for (int updateIter = 0; updateIter < m_options.verboseIter; updateIter++) {
random_shuffle(indexes.begin(), indexes.end());
subExamples.clear();
for (int idy = 0; idy < m_options.batchSize; idy++) {
subExamples.push_back(trainExamples[indexes[idy]]);
}
int curUpdateIter = iter * m_options.verboseIter + updateIter;
cost += m_classifier.process(subExamples, curUpdateIter);
//m_classifier.checkgrads(subExamples, curUpdateIter);
eval.overall_label_count += m_classifier._eval.overall_label_count;
eval.correct_label_count += m_classifier._eval.correct_label_count;
m_classifier.updateParams(m_options.regParameter, m_options.adaAlpha, m_options.adaEps);
}
std::cout << "current iter: " << iter + 1 << ", total iter: " << maxIter << std::endl;
std::cout << "Cost = " << cost << ", SA Correct(%) = " << eval.getAccuracy() << std::endl;
}
if (devNum > 0) {
bCurIterBetter = false;
if (!m_options.outBest.empty())
decodeInstResults.clear();
metric_dev.reset();
for (int idx = 0; idx < devExamples.size(); idx++) {
string result_label;
dtype confidence = predict(devExamples[idx].m_linears, devExamples[idx].m_features, result_label);
devInsts[idx].Evaluate(result_label, metric_dev);
if (!m_options.outBest.empty()) {
curDecodeInst.copyValuesFrom(devInsts[idx]);
curDecodeInst.assignLabel(result_label, confidence);
decodeInstResults.push_back(curDecodeInst);
}
}
metric_dev.print();
if ((!m_options.outBest.empty() && metric_dev.getAccuracy() > bestDIS)) {
m_pipe.outputAllInstances(devFile + m_options.outBest, decodeInstResults);
bCurIterBetter = true;
}
if (testNum > 0) {
if (!m_options.outBest.empty())
decodeInstResults.clear();
metric_test.reset();
for (int idx = 0; idx < testExamples.size(); idx++) {
string result_label;
dtype confidence = predict(testExamples[idx].m_linears, testExamples[idx].m_features, result_label);
testInsts[idx].Evaluate(result_label, metric_test);
if (bCurIterBetter && !m_options.outBest.empty()) {
curDecodeInst.copyValuesFrom(testInsts[idx]);
curDecodeInst.assignLabel(result_label, confidence);
decodeInstResults.push_back(curDecodeInst);
}
}
std::cout << "test:" << std::endl;
metric_test.print();
if ((!m_options.outBest.empty() && bCurIterBetter)) {
m_pipe.outputAllInstances(testFile + m_options.outBest, decodeInstResults);
}
}
for (int idx = 0; idx < otherExamples.size(); idx++) {
std::cout << "processing " << m_options.testFiles[idx] << std::endl;
if (!m_options.outBest.empty())
decodeInstResults.clear();
metric_test.reset();
for (int idy = 0; idy < otherExamples[idx].size(); idy++) {
string result_label;
dtype confidence = predict(otherExamples[idx][idy].m_linears, otherExamples[idx][idy].m_features, result_label);
otherInsts[idx][idy].Evaluate(result_label, metric_test);
if (bCurIterBetter && !m_options.outBest.empty()) {
curDecodeInst.copyValuesFrom(otherInsts[idx][idy]);
curDecodeInst.assignLabel(result_label, confidence);
decodeInstResults.push_back(curDecodeInst);
}
}
std::cout << "test:" << std::endl;
metric_test.print();
if ((!m_options.outBest.empty() && bCurIterBetter)) {
m_pipe.outputAllInstances(m_options.testFiles[idx] + m_options.outBest, decodeInstResults);
}
}
if ((m_options.saveIntermediate && metric_dev.getAccuracy() > bestDIS)) {
if (metric_dev.getAccuracy() > bestDIS) {
std::cout << "Exceeds best previous performance of " << bestDIS << ". Saving model file.." << std::endl;
bestDIS = metric_dev.getAccuracy();
}
writeModelFile(modelFile);
}
}
// Clear gradients
}
if (devNum > 0) {
bCurIterBetter = false;
if (!m_options.outBest.empty())
decodeInstResults.clear();
metric_dev.reset();
for (int idx = 0; idx < devExamples.size(); idx++) {
string result_label;
dtype confidence = predict(devExamples[idx].m_linears, devExamples[idx].m_features, result_label);
devInsts[idx].Evaluate(result_label, metric_dev);
if (!m_options.outBest.empty()) {
curDecodeInst.copyValuesFrom(devInsts[idx]);
curDecodeInst.assignLabel(result_label, confidence);
decodeInstResults.push_back(curDecodeInst);
}
}
metric_dev.print();
if ((!m_options.outBest.empty() && metric_dev.getAccuracy() > bestDIS)) {
m_pipe.outputAllInstances(devFile + m_options.outBest, decodeInstResults);
bCurIterBetter = true;
}
if (testNum > 0) {
if (!m_options.outBest.empty())
decodeInstResults.clear();
metric_test.reset();
for (int idx = 0; idx < testExamples.size(); idx++) {
string result_label;
dtype confidence = predict(testExamples[idx].m_linears, testExamples[idx].m_features, result_label);
testInsts[idx].Evaluate(result_label, metric_test);
if (bCurIterBetter && !m_options.outBest.empty()) {
curDecodeInst.copyValuesFrom(testInsts[idx]);
curDecodeInst.assignLabel(result_label, confidence);
decodeInstResults.push_back(curDecodeInst);
}
}
std::cout << "test:" << std::endl;
metric_test.print();
if ((!m_options.outBest.empty() && bCurIterBetter)) {
m_pipe.outputAllInstances(testFile + m_options.outBest, decodeInstResults);
}
}
for (int idx = 0; idx < otherExamples.size(); idx++) {
std::cout << "processing " << m_options.testFiles[idx] << std::endl;
if (!m_options.outBest.empty())
decodeInstResults.clear();
metric_test.reset();
for (int idy = 0; idy < otherExamples[idx].size(); idy++) {
string result_label;
dtype confidence = predict(otherExamples[idx][idy].m_linears, otherExamples[idx][idy].m_features, result_label);
otherInsts[idx][idy].Evaluate(result_label, metric_test);
if (bCurIterBetter && !m_options.outBest.empty()) {
curDecodeInst.copyValuesFrom(otherInsts[idx][idy]);
curDecodeInst.assignLabel(result_label, confidence);
decodeInstResults.push_back(curDecodeInst);
}
}
std::cout << "test:" << std::endl;
metric_test.print();
if ((!m_options.outBest.empty() && bCurIterBetter)) {
m_pipe.outputAllInstances(m_options.testFiles[idx] + m_options.outBest, decodeInstResults);
}
}
if ((m_options.saveIntermediate && metric_dev.getAccuracy() > bestDIS)) {
if (metric_dev.getAccuracy() > bestDIS) {
std::cout << "Exceeds best previous performance of " << bestDIS << ". Saving model file.." << std::endl;
bestDIS = metric_dev.getAccuracy();
}
writeModelFile(modelFile);
}
} else {
writeModelFile(modelFile);
}
}
bool KNN::train_(ClassificationData &trainingData){
//Clear any previous models
clear();
if( trainingData.getNumSamples() == 0 ){
errorLog << "train_(ClassificationData &trainingData) - Training data has zero samples!" << std::endl;
return false;
}
//Get the ranges of the data
ranges = trainingData.getRanges();
if( useScaling ){
//Scale the training data between 0 and 1
trainingData.scale(0, 1);
}
//Store the number of features, classes and the training data
this->numInputDimensions = trainingData.getNumDimensions();
this->numClasses = trainingData.getNumClasses();
//TODO: In the future need to build a kdtree from the training data to allow better realtime prediction
this->trainingData = trainingData;
//Set the class labels
classLabels.resize( numClasses );
for(UINT k=0; k<numClasses; k++){
classLabels[k] = trainingData.getClassTracker()[k].classLabel;
}
//If we do not need to search for the best K value, then call the sub training function and return the result
if( !searchForBestKValue ){
return train_(trainingData,K);
}
//If we have got this far then we are going to search for the best K value
UINT index = 0;
Float bestAccuracy = 0;
Vector< IndexedDouble > trainingAccuracyLog;
for(UINT k=minKSearchValue; k<=maxKSearchValue; k++){
//Randomly spilt the data and use 80% to train the algorithm and 20% to test it
ClassificationData trainingSet(trainingData);
ClassificationData testSet = trainingSet.split(80,true);
if( !train_(trainingSet, k) ){
errorLog << "Failed to train model for a k value of " << k << std::endl;
}else{
//Compute the classification error
Float accuracy = 0;
for(UINT i=0; i<testSet.getNumSamples(); i++){
VectorFloat sample = testSet[i].getSample();
if( !predict( sample , k) ){
errorLog << "Failed to predict label for test sample with a k value of " << k << std::endl;
return false;
}
if( testSet[i].getClassLabel() == predictedClassLabel ){
accuracy++;
}
}
accuracy = accuracy /Float( testSet.getNumSamples() ) * 100.0;
trainingAccuracyLog.push_back( IndexedDouble(k,accuracy) );
trainingLog << "K:\t" << k << "\tAccuracy:\t" << accuracy << std::endl;
if( accuracy > bestAccuracy ){
bestAccuracy = accuracy;
}
index++;
}
}
if( bestAccuracy > 0 ){
//Sort the training log by value
std::sort(trainingAccuracyLog.begin(),trainingAccuracyLog.end(),IndexedDouble::sortIndexedDoubleByValueDescending);
//Copy the top matching values into a temporary buffer
Vector< IndexedDouble > tempLog;
//Add the first value
tempLog.push_back( trainingAccuracyLog[0] );
//Keep adding values until the value changes
for(UINT i=1; i<trainingAccuracyLog.size(); i++){
if( trainingAccuracyLog[i].value == tempLog[0].value ){
tempLog.push_back( trainingAccuracyLog[i] );
}else break;
}
//Sort the temp values by index (the index is the K value so we want to get the minimum K value with the maximum accuracy)
std::sort(tempLog.begin(),tempLog.end(),IndexedDouble::sortIndexedDoubleByIndexAscending);
trainingLog << "Best K Value: " << tempLog[0].index << "\tAccuracy:\t" << tempLog[0].value << std::endl;
//Use the minimum index, this should give us the best accuracy with the minimum K value
//We now need to train the model again to make sure all the training metrics are computed correctly
return train_(trainingData,tempLog[0].index);
}
return false;
}
//返回预测的角速度
double RobotTracker::angular_velocity(double time)
{
Matrix x = predict(time);
return x.e(5,0);
}
void mexFunction( int nlhs, mxArray *plhs[],
int nrhs, const mxArray *prhs[] )
{
int prob_estimate_flag = 0;
struct svm_model *model;
if(nrhs > 4 || nrhs < 3)
{
exit_with_help();
fake_answer(plhs);
return;
}
if(!mxIsDouble(prhs[0]) || !mxIsDouble(prhs[1])) {
mexPrintf("Error: label vector and instance matrix must be double\n");
fake_answer(plhs);
return;
}
if(mxIsStruct(prhs[2]))
{
const char *error_msg;
/* parse options */
if(nrhs==4)
{
int i, argc = 1;
char cmd[CMD_LEN], *argv[CMD_LEN/2];
/* put options in argv[] */
mxGetString(prhs[3], cmd, mxGetN(prhs[3]) + 1);
if((argv[argc] = strtok(cmd, " ")) != NULL)
while((argv[++argc] = strtok(NULL, " ")) != NULL)
;
for(i=1;i<argc;i++)
{
if(argv[i][0] != '-') break;
if(++i>=argc)
{
exit_with_help();
fake_answer(plhs);
return;
}
switch(argv[i-1][1])
{
case 'b':
prob_estimate_flag = atoi(argv[i]);
break;
default:
mexPrintf("Unknown option: -%c\n", argv[i-1][1]);
exit_with_help();
fake_answer(plhs);
return;
}
}
}
model = matlab_matrix_to_model(prhs[2], &error_msg);
if (model == NULL)
{
mexPrintf("Error: can't read model: %s\n", error_msg);
fake_answer(plhs);
return;
}
if(prob_estimate_flag)
{
if(svm_check_probability_model(model)==0)
{
mexPrintf("Model does not support probabiliy estimates\n");
fake_answer(plhs);
svm_free_and_destroy_model(&model);
return;
}
}
else
{
if(svm_check_probability_model(model)!=0)
printf("Model supports probability estimates, but disabled in predicton.\n");
}
predict(plhs, prhs, model, prob_estimate_flag);
/* destroy model */
svm_free_and_destroy_model(&model);
}
else
{
mexPrintf("model file should be a struct array\n");
fake_answer(plhs);
}
return;
}
void RTFusionKalman4::newIMUData(RTIMU_DATA& data, const RTIMUSettings *settings)
{
if (m_enableGyro)
m_gyro = data.gyro;
else
m_gyro = RTVector3();
m_accel = data.accel;
m_compass = data.compass;
m_compassValid = data.compassValid;
if (m_firstTime) {
m_lastFusionTime = data.timestamp;
calculatePose(m_accel, m_compass, settings->m_compassAdjDeclination);
m_Fk.fill(0);
// init covariance matrix to something
m_Pkk.fill(0);
for (int i = 0; i < 4; i++)
for (int j = 0; j < 4; j++)
m_Pkk.setVal(i,j, 0.5);
// initialize the observation model Hk
// Note: since the model is the state vector, this is an identity matrix so it won't be used
// initialize the poses
m_stateQ.fromEuler(m_measuredPose);
m_fusionQPose = m_stateQ;
m_fusionPose = m_measuredPose;
m_firstTime = false;
} else {
m_timeDelta = (RTFLOAT)(data.timestamp - m_lastFusionTime) / (RTFLOAT)1000000;
m_lastFusionTime = data.timestamp;
if (m_timeDelta <= 0)
return;
if (m_debug) {
HAL_INFO("\n------\n");
HAL_INFO1("IMU update delta time: %f\n", m_timeDelta);
}
calculatePose(data.accel, data.compass, settings->m_compassAdjDeclination);
predict();
update();
m_stateQ.toEuler(m_fusionPose);
m_fusionQPose = m_stateQ;
if (m_debug | settings->m_fusionDebug) {
HAL_INFO(RTMath::displayRadians("Measured pose", m_measuredPose));
HAL_INFO(RTMath::displayRadians("Kalman pose", m_fusionPose));
HAL_INFO(RTMath::displayRadians("Measured quat", m_measuredPose));
HAL_INFO(RTMath::display("Kalman quat", m_stateQ));
HAL_INFO(RTMath::display("Error quat", m_stateQError));
}
}
data.fusionPoseValid = true;
data.fusionQPoseValid = true;
data.fusionPose = m_fusionPose;
data.fusionQPose = m_fusionQPose;
}
void FunctionApproximator::train(const MatrixXd& inputs, const MatrixXd& targets, string save_directory, bool overwrite)
{
train(inputs,targets);
if (save_directory.empty())
return;
if (!isTrained())
return;
if (getExpectedInputDim()<3)
{
VectorXd min = inputs.colwise().minCoeff();
VectorXd max = inputs.colwise().maxCoeff();
int n_samples_per_dim = 100;
if (getExpectedInputDim()==2) n_samples_per_dim = 40;
VectorXi n_samples_per_dim_vec = VectorXi::Constant(getExpectedInputDim(),n_samples_per_dim);
MatrixXd inputs_grid;
FunctionApproximator::generateInputsGrid(min, max, n_samples_per_dim_vec, inputs_grid);
MatrixXd outputs_grid(inputs_grid.rows(),1);
predict(inputs_grid,outputs_grid);
saveMatrix(save_directory,"n_samples_per_dim.txt",n_samples_per_dim_vec,overwrite);
saveMatrix(save_directory,"inputs_grid.txt",inputs_grid,overwrite);
saveMatrix(save_directory,"outputs_grid.txt",outputs_grid,overwrite);
MatrixXd variances_grid;
predictVariance(inputs_grid,variances_grid);
if (!variances_grid.size()==0)
{
variances_grid = variances_grid.array().sqrt();
saveMatrix(save_directory,"variances_grid.txt",variances_grid,overwrite);
}
model_parameters_->saveGridData(min, max, n_samples_per_dim_vec, save_directory, overwrite);
}
MatrixXd outputs;
predict(inputs,outputs);
saveMatrix(save_directory,"inputs.txt",inputs,overwrite);
saveMatrix(save_directory,"targets.txt",targets,overwrite);
saveMatrix(save_directory,"outputs.txt",outputs,overwrite);
string filename = save_directory+"/plotdata.py";
ofstream outfile;
outfile.open(filename.c_str());
if (!outfile.is_open())
{
cerr << __FILE__ << ":" << __LINE__ << ":";
cerr << "Could not open file " << filename << " for writing." << endl;
}
else
{
// Python code generation in C++. Rock 'n' roll! ;-)
if (inputs.cols()==2) {
outfile << "from mpl_toolkits.mplot3d import Axes3D \n";
}
outfile << "import numpy \n";
outfile << "import matplotlib.pyplot as plt \n";
outfile << "directory = '" << save_directory << "' \n";
outfile << "inputs = numpy.loadtxt(directory+'/inputs.txt') \n";
outfile << "targets = numpy.loadtxt(directory+'/targets.txt') \n";
outfile << "outputs = numpy.loadtxt(directory+'/outputs.txt') \n";
outfile << "fig = plt.figure() \n";
if (inputs.cols()==2) {
outfile << "ax = Axes3D(fig) \n";
outfile << "ax.plot(inputs[:,0],inputs[:,1],targets, '.', label='targets',color='black') \n";
outfile << "ax.plot(inputs[:,0],inputs[:,1],outputs, '.', label='predictions',color='red')\n";
outfile << "ax.set_xlabel('input_1'); ax.set_ylabel('input_2'); ax.set_zlabel('output') \n";
outfile << "ax.legend(loc='lower right') \n";
} else {
outfile << "plt.plot(inputs,targets, '.', label='targets',color='black') \n";
outfile << "plt.plot(inputs,outputs, '.', label='predictions',color='red') \n";
outfile << "plt.xlabel('input'); plt.ylabel('output'); \n";
outfile << "plt.legend(loc='lower right') \n";
}
outfile << "plt.show() \n";
outfile << endl;
outfile.close();
//cout << " ______________________________________________________________" << endl;
//cout << " | Plot saved data with:" << " 'python " << filename << "'." << endl;
//cout << " |______________________________________________________________" << endl;
}
}
void Recommender::cache_recommendations(set<int>& users,
set<int>& articles,
bool startFresh)
{
time_t _start_time = time(NULL);
int terminator = -10;
int num_total_recs = 0;
for(set<int>::iterator user = users.begin();
user != users.end() && terminator--; user++) {
char buffer[50];
sprintf(buffer, "recommend_%d", (*user));
User u;
u.clear();
if (!startFresh)
u.decachify(buffer);
time_t start_time = time(NULL);
int n_attempts = 0, n_attempts_max = 10;
do {
int rec_count = 0;
for(set<int>::iterator art = articles.begin();
art != articles.end(); art++) {
if(!this->storeModdedArticles &&
mods[(*user)].has_key(*art)) continue;
float q = predict(*user, *art, n_attempts != 0 && art == articles.begin());
int a = (*art);
if(q != 0) {
rec_count++;
u.add(a, q);
}
}
if(!rec_count && articles.size() > 100) {
cout << " +++ Resorting to cluster behavior" << endl;
double fudge0 = fudge;
fudge = 1;
for(set<int>::iterator art = articles.begin();
art != articles.end(); art++) {
if(mods[(*user)].has_key(*art)) continue;
float q = predict(*user, *art);
int a = (*art);
if(q != 0) {
rec_count++;
u.add(a, q);
}
}
fudge = fudge0;
}
num_total_recs += rec_count;
if(rec_count > 20) {
cout << " --> " << (*user)
<< ": \x1b[32m\x1b[1mGenerated " << rec_count << " additional recommendations\x1b[0m"
<< endl;
break;
}
else {
cout << " --> \x1b[33m\x1b[1mFAILED TO GENERATE RECOMMENDATIONS (only " << rec_count << ")!!!\x1b[0m" << endl;
cout << " Available mods: " << mods[(*user)].size() << endl;
char buffer[100];
if(mods[(*user)].size() > 5) {
cout << " --> \x1b[31m\x1b[1mFAILED TO GENERATE RECOMMENDATIONS (in the bad way)!!!\x1b[0m" << endl;
}
}
} while(++n_attempts < n_attempts_max);
if(u.size())
u.memcachify(buffer);
// recs[*user].memcachify(buffer);
cout << "\tFinished user " << (*user) << " ("
<< difftime(time(NULL), start_time) << " sec, "
<< articles.size() << " arts)" << endl;
}
cout << "Total recommendation time: " << difftime(time(NULL), _start_time)
<< " / " << users.size() << " users * "
<< articles.size() << " articles \n\t --> "
<< num_total_recs << " recs (@ "
<< 1000./float(num_total_recs)*float(difftime(time(NULL), _start_time)) << " ms each)" << endl;
// fclose(toFile(fopen("user.data", "wb")));
}
void InertiaSensorFilter::update(OrientationData& orientationData,
const InertiaSensorData& theInertiaSensorData,
const SensorData& theSensorData,
const RobotModel& theRobotModel,
const FrameInfo& theFrameInfo,
const MotionInfo& theMotionInfo,
const WalkingEngineOutput& theWalkingEngineOutput)
{
// MODIFY("module:InertiaSensorFilter:parameters", p);
//
// DECLARE_PLOT("module:InertiaSensorFilter:expectedGyroX");
// DECLARE_PLOT("module:InertiaSensorFilter:gyroX");
// DECLARE_PLOT("module:InertiaSensorFilter:expectedGyroY");
// DECLARE_PLOT("module:InertiaSensorFilter:gyroY");
// DECLARE_PLOT("module:InertiaSensorFilter:expectedAccX");
// DECLARE_PLOT("module:InertiaSensorFilter:accX");
// DECLARE_PLOT("module:InertiaSensorFilter:expectedAccY");
// DECLARE_PLOT("module:InertiaSensorFilter:accY");
// DECLARE_PLOT("module:InertiaSensorFilter:expectedAccZ");
// DECLARE_PLOT("module:InertiaSensorFilter:accZ");
// check whether the filter shall be reset
if(!lastTime || theFrameInfo.time <= lastTime)
{
if(theFrameInfo.time == lastTime)
return; // weird log file replaying?
x = State<>();
cov = p.processCov;
lastLeftFoot = lastRightFoot = Pose3D();
lastTime = theFrameInfo.time - (unsigned int)(theFrameInfo.cycleTime * 1000.f);
}
// get foot positions
const Pose3D& leftFoot(theRobotModel.limbs[MassCalibration::footLeft]);
const Pose3D& rightFoot(theRobotModel.limbs[MassCalibration::footRight]);
const Pose3D leftFootInvert(leftFoot.invert());
const Pose3D rightFootInvert(rightFoot.invert());
// calculate rotation and position offset using the robot model (joint data)
const Pose3D leftOffset(lastLeftFoot.translation.z != 0.f ? Pose3D(lastLeftFoot).conc(leftFootInvert) : Pose3D());
const Pose3D rightOffset(lastRightFoot.translation.z != 0.f ? Pose3D(lastRightFoot).conc(rightFootInvert) : Pose3D());
// detect the foot that is on ground
bool useLeft = true;
if(theMotionInfo.motion == MotionRequest::walk && theWalkingEngineOutput.speed.translation.x != 0)
useLeft = theWalkingEngineOutput.speed.translation.x > 0 ?
(leftOffset.translation.x > rightOffset.translation.x) :
(leftOffset.translation.x < rightOffset.translation.x);
else
{
Pose3D left(x.rotation);
Pose3D right(x.rotation);
left.conc(leftFoot);
right.conc(rightFoot);
useLeft = left.translation.z < right.translation.z;
}
// calculate velocity
Vector3<> calcVelocity, lastCalcVelocity;
float timeScale = 1.f / (float(theFrameInfo.time - lastTime) * 0.001f);
calcVelocity = useLeft ? leftOffset.translation : rightOffset.translation;
calcVelocity *= timeScale * 0.001f; // => m/s
// update the filter
timeScale = float(theFrameInfo.time - lastTime) * 0.001f;
predict(theInertiaSensorData.gyro.x != InertiaSensorData::off ?
RotationMatrix(Vector3<>(theInertiaSensorData.gyro.x * timeScale, theInertiaSensorData.gyro.y * timeScale, 0)) :
(useLeft ? leftOffset.rotation : rightOffset.rotation));
// insert calculated rotation
if(theInertiaSensorData.acc.x != InertiaSensorData::off)
safeRawAngle = Vector2<>(theSensorData.data[SensorData::angleX], theSensorData.data[SensorData::angleY]);
if((theMotionInfo.motion == MotionRequest::walk || theMotionInfo.motion == MotionRequest::stand ||
(theMotionInfo.motion == MotionRequest::specialAction && theMotionInfo.specialActionRequest.specialAction == SpecialActionRequest::sitDownKeeper)) &&
abs(safeRawAngle.x) < p.calculatedAccLimit.x && abs(safeRawAngle.y) < p.calculatedAccLimit.y)
{
const RotationMatrix& usedRotation(useLeft ? leftFootInvert.rotation : rightFootInvert.rotation);
RotationMatrix calculatedRotation(Vector3<>(
atan2(usedRotation.c1.z, usedRotation.c2.z),
atan2(-usedRotation.c0.z, usedRotation.c2.z), 0.f));
Vector3<> accGravOnly(calculatedRotation.c0.z, calculatedRotation.c1.z, calculatedRotation.c2.z);
accGravOnly *= -9.80665f;
readingUpdate(accGravOnly);
}
else // insert acceleration sensor values
{
if(theInertiaSensorData.acc.x != InertiaSensorData::off)
readingUpdate(theInertiaSensorData.acc);
}
// fill the representation
orientationData.orientation = Vector2<>(
atan2(x.rotation.c1.z, x.rotation.c2.z),
atan2(-x.rotation.c0.z, x.rotation.c2.z));
// this removes any kind of z-rotation from internal rotation
if(orientationData.orientation.squareAbs() < 0.04f * 0.04f)
x.rotation = RotationMatrix(Vector3<>(orientationData.orientation.x, orientationData.orientation.y, 0.f));
orientationData.velocity = calcVelocity;
// store some data for the next iteration
lastLeftFoot = leftFoot;
lastRightFoot = rightFoot;
lastTime = theFrameInfo.time;
// plots
// PLOT("module:InertiaSensorFilter:orientationX", orientationData.orientation.x);
// PLOT("module:InertiaSensorFilter:orientationY", orientationData.orientation.y);
// PLOT("module:InertiaSensorFilter:velocityX", orientationData.velocity.x);
// PLOT("module:InertiaSensorFilter:velocityY", orientationData.velocity.y);
// PLOT("module:InertiaSensorFilter:velocityZ", orientationData.velocity.z);
}
bool KNN::train(LabelledClassificationData &trainingData){
if( !searchForBestKValue ){
return train_(trainingData,K);
}
UINT bestIndex = 0;
UINT index = 0;
double bestAccuracy = 0;
vector< IndexedDouble > trainingAccuracyLog;
for(UINT k=minKSearchValue; k<=maxKSearchValue; k++){
//Randomly spilt the data and use 80% to train the algorithm and 20% to test it
LabelledClassificationData trainingSet(trainingData);
LabelledClassificationData testSet = trainingSet.partition(80,true);
if( !train_(trainingSet, k) ){
errorLog << "Failed to train model for a k value of " << k << endl;
}else{
//Compute the classification error
double accuracy = 0;
for(UINT i=0; i<testSet.getNumSamples(); i++){
vector< double > sample = testSet[i].getSample();
if( !predict( sample ) ){
errorLog << "Failed to predict label for test sample with a k value of " << k << endl;
return false;
}
if( testSet[i].getClassLabel() == predictedClassLabel ){
accuracy++;
}
}
accuracy = accuracy /double( testSet.getNumSamples() ) * 100.0;
trainingAccuracyLog.push_back( IndexedDouble(k,accuracy) );
trainingLog << "K:\t" << k << "\tAccuracy:\t" << accuracy << endl;
if( accuracy > bestAccuracy ){
bestIndex = index;
bestAccuracy = accuracy;
}
index++;
}
}
if( bestAccuracy > 0 ){
//Sort the training log by value
std::sort(trainingAccuracyLog.begin(),trainingAccuracyLog.end(),IndexedDouble::sortIndexedDoubleByValueAscending);
//Copy the top matching values into a temporary buffer
vector< IndexedDouble > tempLog;
//Add the first value
tempLog.push_back( trainingAccuracyLog[0] );
//Keep adding values until the value changes
for(UINT i=1; i<trainingAccuracyLog.size(); i++){
if( trainingAccuracyLog[i].value == tempLog[0].value ){
tempLog.push_back( trainingAccuracyLog[i] );
}else break;
}
//Sort the temp values by index
std::sort(tempLog.begin(),tempLog.end(),IndexedDouble::sortIndexedDoubleByIndexDescending);
trainingLog << "Best K Value: " << tempLog[0].index << "\tAccuracy:\t" << tempLog[0].value << endl;
//Use the minimum index, this should give us the best accuracy with the minimum K value
return train_(trainingData,tempLog[0].index);
}
return false;
}
/* Use linear form for r, computes inverse model using inverse_Fx */
Float predict (Linear_invertable_predict_model& f)
/* Use linear form for r, and use inv.Fx from invertible model */
{
return predict(f, f.inv.Fx, true);
}
int main(int argc, char **argv)
{
FILE *input, *output;
int i;
// parse options
for(i=1;i<argc;i++)
{
if(argv[i][0] != '-') break;
++i;
switch(argv[i-1][1])
{
case 'b':
predict_probability = atoi(argv[i]);
break;
case 'q':
info = &print_null;
i--;
break;
default:
fprintf(stderr,"Unknown option: -%c\n", argv[i-1][1]);
exit_with_help();
}
}
if(i>=argc-2)
exit_with_help();
input = fopen(argv[i],"r");
if(input == NULL)
{
fprintf(stderr,"can't open input file %s\n",argv[i]);
exit(1);
}
output = fopen(argv[i+2],"w");
if(output == NULL)
{
fprintf(stderr,"can't open output file %s\n",argv[i+2]);
exit(1);
}
if((model=svm_load_model(argv[i+1]))==0)
{
fprintf(stderr,"can't open model file %s\n",argv[i+1]);
exit(1);
}
#ifdef _DENSE_REP
x.dim = 0;
x.values = (double*) malloc( max_nr_attr*sizeof(double) );
#else
x = (struct svm_node *) malloc(max_nr_attr*sizeof(struct svm_node));
#endif
if(predict_probability)
{
if(svm_check_probability_model(model)==0)
{
fprintf(stderr,"Model does not support probabiliy estimates\n");
exit(1);
}
}
else
{
if(svm_check_probability_model(model)!=0)
info("Model supports probability estimates, but disabled in prediction.\n");
}
predict(input,output);
svm_free_and_destroy_model(&model);
#ifdef CL_SVM
svm_teardown_prediction();
#endif
#ifdef _DENSE_REP
free( x.values );
#else
free(x);
#endif
free(line);
fclose(input);
fclose(output);
return 0;
}
void ParticleFilter::filter(double camX, double camY, double camD, double* resX, double* resY, double* resD){
resample();
predict();
weight(camX, camY, camD);
measure(resX, resY, resD);
}
