// Returns a new CubicSpline setting array lengths to n and all array values to zero with natural spline default
// Primarily for use in BiCubicSpline
public virtual CubicSpline zero(int n)
{
if (n < 3)
throw new System.ArgumentException("A minimum of three data points is needed");
CubicSpline aa = new CubicSpline(n);
return aa;
}
// Create a one dimensional array of cubic spline objects of length n each of array length m
// Primarily for use in BiCubicSpline
public virtual CubicSpline[] oneDarray(int n, int m)
{
if (m < 3)
throw new System.ArgumentException("A minimum of three data points is needed");
CubicSpline[] a = new CubicSpline[n];
for (int i = 0; i < n; i++)
{
a[i] = zero(m);
}
return a;
}
public static bool GenerateFeatureVector(double lastTimeStamp)
{
// Generate a feature vector only, if the next window end has
// passed based on the configuration parameters (window size and overlap)
// otherwise return false
if (lastTimeStamp < Extractor.next_window_end)
return false;
// the last time stamp is more than the next expected window end
// At this point, we have a complete window ready for feature calculation
//compute the boundaries for the current window
double window_start_time = lastTimeStamp - Extractor.dconfiguration.WindowTime;
double window_end_time = lastTimeStamp;
double current_time = window_end_time;
//compute the end of the next overlapping window
next_window_end = window_end_time + (Extractor.dconfiguration.WindowTime * Extractor.dconfiguration.WindowOverlap);
#region sensors window grabbing and interpolation
// Go through each sensor and extract the collected data within
// the current time window
for (int j = 0; (j < Extractor.extractorSensorCount); j++)
{
// Check that the previous sensor in the loop did not report
// deteriorated quality for its data
#region sensors window quality
if (isAcceptableLossRate == false)
break;
// check if earlier axes reported excessive consecutive loss of data frames
if (isAcceptableConsecutiveLoss == false)
{
Extractor.discardedConsecutiveLossWindows++;
break;
}
#endregion sensors window quality
// determine the base index for the current sensor in data array, each sensor has 4 rows (x,y,z,timestamp)
int sensor_index = j * 4;
int time_index = sensor_index + 3;
// determine the last read data sample for the current sensor
// by looking at its index
int last_sample = 0;
if (y_index[j] == 0)
last_sample = Extractor.EXPECTED_WINDOW_SIZES[j] - 1;
else
last_sample = y_index[j] - 1;
// int total_data_points = 0, distinct_data_points = 0;
int distinct_data_points = 0;
//Grab the readings for each axis of a sensor and smoothen it
#region sensor window grabbing and interpolation
// Go through each axis of the current sensor and smooth
// it using the cubic spline
for (int axes_num = 0; (axes_num < 3); axes_num++)
{
//calculate the exact index based on the
// base sensor index and the axis number
int axes_index = sensor_index + axes_num; //for data array
int interpolated_axes_index = j * 3 + axes_num; //for interpolated data array
// create 2 arrays to store x and y values for the cubic spline
// it is sufficient to have an array of expected sampling rate window size
// for 3 mites that would be 180/60
double[] xvals = new double[Extractor.EXPECTED_WINDOW_SIZES[j]];
double[] yvals = new double[Extractor.EXPECTED_WINDOW_SIZES[j]];
//point to the last sample that was read and get its time
int sample_index = last_sample;
current_time = data[time_index][sample_index];
//copy samples in the last time window
// total_data_points = 0;
distinct_data_points = 0;
// double previous_time = 0;
//Grab the values for a specific sensor axes between
//window start and window end
#region window grabbing
// Start going back from the current time (window end) till the start of the window
// without exceeding the expected sampling rate and fill in the data in the signal
// value for the axis in yvals and the relative time value from the window start
while ((current_time >= window_start_time) && (current_time <= window_end_time)
&& (distinct_data_points < Extractor.EXPECTED_WINDOW_SIZES[j]))
{
//some time stamps from the mites are identical
// for interpolation that will cause an error
// simply take the first value for a time point and ignore the
// rest, another strategy would be to average over these values
//if (current_time == previous_time)
//{
// //Get the time of the previous sample and skip the sample
// if (sample_index == 0)
// sample_index = EXPECTED_WINDOW_SIZE - 1;
// else
// sample_index--;
// current_time = data[time_index][sample_index];
// total_data_points++;
// continue;
//}
// Quality Control
// check the time between consecutive data frames and make sure it does
// not exceed maximum_consecutive_loss, do not do that for the first
// entry of the window
// Not suitable for the phone due to time resolution
//if (distinct_data_points > 0)
//{
// int consecutive_lost_packets = (int)((previous_time - current_time) / EXPECTED_SAMPLES_SPACING);
// if (consecutive_lost_packets > Extractor.dconfiguration.MaximumConsecutiveFrameLoss)
// {
// Extractor.discardedConsecutiveLossWindows++;
// unacceptable_consecutive_window_loss_count++;
// isAcceptableConsecutiveLoss = false;
// break;
// }
//}
//some time stamps from the mites are identical
// for interpolation that will cause an error
// simply take the first value for a time point and ignore the
// rest, another strategy would be to average over these values
// we decided, we will spread them out evenly as long as
// no excessive loss is experienced
// done down in the code
xvals[distinct_data_points] = (int)(current_time - window_start_time);
//signal value for the current sample and current axis.
yvals[distinct_data_points] = data[axes_index][sample_index];
//Point to the previous sample in the current window
if (sample_index == 0)
sample_index = Extractor.EXPECTED_WINDOW_SIZES[j] - 1;
else
sample_index--;
//store the previous sample time
// previous_time = current_time;
//Get the time of the new sample
current_time = data[time_index][sample_index];
//Point to the next entry in the interpolation array
distinct_data_points++;
// total_data_points++;
}
#endregion window grabbing
//Check if the captured window has acceptable loss rate
#region window quality checks
//Do not proceed if there was excessive consecutive loss of data frames
if (isAcceptableConsecutiveLoss == false)
break;
// all data for a specific sensor axis for the current window are stored
// in xvals and yvals
// check if the data is admissible for feature calculation according to the following
// criteria:
// 1- total lost data frames are within the loss rate
// 2- the number of consecutive lost packets is within our maximum_consecutive_loss parameter
if (distinct_data_points < Extractor.EXPECTED_GOOD_SAMPLING_RATES[j]) //discard this whole window of data
{
Extractor.discardedLossRateWindows++;
isAcceptableLossRate = false;
unacceptable_window_count++;
break;
}
#endregion window quality checks
//smoothen the axis values and store them in interpolated data array
#region window interpolation
//create 2 arrays with the exact size of the data points for interpolation
double[] admissible_xvals = new double[distinct_data_points];
double[] admissible_yvals = new double[distinct_data_points];
double expectedSpacing = Extractor.dconfiguration.WindowTime / (double)distinct_data_points;
double startTime = 0.0;
for (int k = 0; (k < distinct_data_points); k++)
{
admissible_xvals[k] = startTime+k*expectedSpacing;//xvals[distinct_data_points - k - 1];
admissible_yvals[k] = yvals[distinct_data_points - k - 1];
}
//if (spacingtws == null)
// {
// spacingtws = new TextWriter[Extractor.sannotation.Sensors.Count];
// for (int i = 0; (i < Extractor.sannotation.Sensors.Count); i++)
// {
// spacingtws[i] = new StreamWriter("\\test\\" + "SensorSpacing" + i + ".txt");
// }
// }
// smooth it using a cubic spline
CubicSpline cs= new CubicSpline(admissible_xvals, admissible_yvals);
//startval = xvals[distinct_data_points - 1].ToString("00.00");
// shrink or expand the data window using interpolation
for (int k = 0; (k < INTERPOLATED_SAMPLING_RATE_PER_WINDOW); k++)
{
interpolated_data[interpolated_axes_index][k] = cs.interpolate(k * INTERPOLATED_SAMPLES_SPACING);
//check that the intrepolated values make sense.
//if ((interpolated_data[interpolated_axes_index][k] <= 0) || (interpolated_data[interpolated_axes_index][k] > 1024))
if ((interpolated_data[interpolated_axes_index][k] <= -6000) || (interpolated_data[interpolated_axes_index][k] > 6000))
{
// errorFlag = 1;
return false;
}
}
#endregion window interpolation
}
#endregion sensor window grabbing and interpolation
//spacingtws[j].WriteLine(startval);
}
#endregion sensors window grabbing and interpolation
//If the data is of acceptable quality, calculate the features
#region Calculate Feature Vector
if ((isAcceptableLossRate == true) && (isAcceptableConsecutiveLoss == true))
{
//Extract the features from the interpolated data
//Extractor.Extract(interpolated_data);
Extractor.Extract(interpolated_data);
//Output the data to the ARFF file
// tw.WriteLine(Extractor.toString() + "," + current_activity);
//num_feature_windows++;
// make sure the values of the features make sense... for example distance can
// only be -1024---- or so
return true;
}
else //the window is of poor quality, reinitialize and continue
{
isAcceptableConsecutiveLoss = true;
isAcceptableLossRate = true;
//errorFlag = 2;
return false;
}
#endregion Calculate Feature Vector
}