def formatTrainingSet(self, setClass):
# recording params
Ifilename = "training_data_" + setClass + ".txt"
Ofilename = "training_data_" + setClass + ".yaml"
# open, read and process sets
try:
Ifile = open(Ifilename,'r')
except IOError:
print("Data File does not exist. Please run Training Program Before Running")
raise IOError
arr = []
fullarr = []
for line in Ifile:
val = line.split(',')
arr = []
for i in val:
arr.append(float(i))
fullarr.append(arr)
test = numpy.array(fullarr, dtype ="float")
test = numpy.transpose(test)
io.saveYaml(Ofilename, test)
Ifile.close()
print("File successfully formatted")
def formatTrainingSet(self, setClass):
# recording params
Ifilename = "training_data_" + setClass + ".txt"
Ofilename = "training_data_" + setClass + ".yaml"
# open, read and process sets
try:
Ifile = open(Ifilename,'r')
except IOError:
print("Data File does not exist. Please run Training Program Before Running")
raise IOError
arr = []
fullarr = []
for line in Ifile:
val = line.split(',')
arr = []
for i in val:
arr.append(float(i))
fullarr.append(arr)
test = numpy.array(fullarr, dtype ="float")
test = numpy.transpose(test)
io.saveYaml(Ofilename, test)
Ifile.close()
print("File successfully formatted")
def reloadPCAParams(self):
print ".......................... RELOAD PARAMS ................................"
# load the setClassAverages
self.setClassAverages = io.fetchYaml('pca_scores.yaml')
# load the pca_params
self.pca_params = io.fetchYaml('pca_params.yaml')
# clear serial buffer
timeI = time.time()
while time.time() - timeI < 3.5:
raw_data = self.ser.readline()
# graph debug file
self.error = {'left_nod': [], 'right_nod': []}
io.saveYaml('error_graphs_data.yaml', self.error)
def reloadPCAParams(self):
print ".......................... RELOAD PARAMS ................................"
# load the setClassAverages
self.setClassAverages = io.fetchYaml('pca_scores.yaml')
# load the pca_params
self.pca_params = io.fetchYaml('pca_params.yaml')
# clear serial buffer
timeI = time.time()
while time.time() - timeI < 3.5:
raw_data = self.ser.readline()
# graph debug file
self.error = {'left_nod': [], 'right_nod': []}
io.saveYaml('error_graphs_data.yaml', self.error)
def __init__(self, baudrate=57600, portname="/dev/ttyACM0", numberOfSets=10, window=80):
self.baudrate = baudrate
self.portname = portname
self.numberOfSets = numberOfSets
self.sensorHistory = []
self.window = window
# load the setClassAverages
self.setClassAverages = io.fetchYaml('pca_scores.yaml')
# load the pca_params
self.pca_params = io.fetchYaml('pca_params.yaml')
# connect to available devices
self.ser = io.connectToAvailablePort(baudrate=self.baudrate, portName=self.portname, debug=True)
# clear serial buffer
timeI = time.time()
while time.time() - timeI < 3.5:
raw_data = self.ser.readline()
# graph debug file
self.error = {'left_nod': [], 'right_nod': []}
io.saveYaml('error_graphs_data.yaml', self.error)
def trainModel(self, setClasses=['left_nod', 'right_nod']):
# ======== read in training data ========
trainedData = []
trainedDataDims = []
trainedDataDatapoints = []
for setClassIndex in range(0, len(setClasses)):
setClassFilename = 'training_data_' + str(setClasses[setClassIndex]) + '.yaml'
trainedData.append(io.fetchYaml(setClassFilename))
(d, dp) = trainedData[setClassIndex].shape
trainedDataDims.append(d)
trainedDataDatapoints.append(dp)
# check input dimension consistency across training sets
inputDims = trainedDataDims[0]
for d in trainedDataDims:
if inputDims != d:
print "Number of dimensions in the training sets do not match"
sys.exit()
# ======== concatenate training files ========
pcaInput = []
for setClassIndex in range(0, len(trainedData)):
for colIndex in range(0, trainedDataDatapoints[setClassIndex]):
dp = trainedData[setClassIndex][:, colIndex].tolist()
pcaInput.append(dp)
pcaInput = np.transpose(np.array(pcaInput))
# ======== do PCA ========
coeff, score, latent = pca.princomp(pcaInput.T, self.dims)
varience_covered = np.sum(latent[0:self.dims]) / np.sum(latent)
startingIndex = 0
endingIndex = 0
setClassScores = {}
for setClassIndex in range(0, len(setClasses)):
startingIndex = endingIndex
endingIndex = endingIndex + trainedDataDatapoints[setClassIndex]
setClassScores[setClasses[setClassIndex]] = score[:, startingIndex:endingIndex-1]
pca_params ={}
pca_params['coeff'] = coeff
pca_params['latent'] = latent
pca_params['dims'] = self.dims
pca_params['window'] = self.window
io.saveYaml('pca_params.yaml', pca_params)
# ======== find average of training sets ========
# find peaks in left score
setClassAverages = {}
for setClassIndex in range(0, len(setClasses)):
setClassAverage = []
setClassScore = setClassScores[setClasses[setClassIndex]]
for peakNumber in range(0, self.numberOfNods):
(d, dp) = setClassScore.shape
peakLocation = np.argmax(np.absolute(setClassScore))%dp
columnsInWindow = range(peakLocation-self.window/2, peakLocation+self.window/2)
setClassAverage.append(setClassScore[:, columnsInWindow])
setClassScore = np.delete(setClassScore, columnsInWindow, 1)
# compute average
setClassAverages[setClasses[setClassIndex]] = np.mean(np.array(setClassAverage), axis=0)
# normalize
for setClass in setClassAverages:
a = setClassAverages[setClass]
row_sums = np.absolute(a).sum(axis=1)
setClassAverages[setClass] = a / row_sums[:, np.newaxis]
# save scores
io.saveYaml('pca_scores.yaml', setClassAverages)
# plot all scores
print "plotting"
for setClass in setClassAverages:
fig = plt.figure()
ax = fig.add_subplot(111)
ax.plot(np.transpose(setClassAverages[setClass]))
filename = 'graph_' + str(setClass) + '.png'
plt.savefig(filename)
def trainModel(self, setClasses=['left_nod', 'right_nod']):
# ======== read in training data ========
trainedData = []
trainedDataDims = []
trainedDataDatapoints = []
for setClassIndex in range(0, len(setClasses)):
setClassFilename = 'training_data_' + str(setClasses[setClassIndex]) + '.yaml'
trainedData.append(io.fetchYaml(setClassFilename))
(d, dp) = trainedData[setClassIndex].shape
trainedDataDims.append(d)
trainedDataDatapoints.append(dp)
# check input dimension consistency across training sets
inputDims = trainedDataDims[0]
for d in trainedDataDims:
if inputDims != d:
print "Number of dimensions in the training sets do not match"
sys.exit()
# ======== concatenate training files ========
pcaInput = []
for setClassIndex in range(0, len(trainedData)):
for colIndex in range(0, trainedDataDatapoints[setClassIndex]):
dp = trainedData[setClassIndex][:, colIndex].tolist()
pcaInput.append(dp)
pcaInput = np.transpose(np.array(pcaInput))
# ======== do PCA ========
coeff, score, latent = pca.princomp(pcaInput.T, self.dims)
varience_covered = np.sum(latent[0:self.dims]) / np.sum(latent)
startingIndex = 0
endingIndex = 0
setClassScores = {}
for setClassIndex in range(0, len(setClasses)):
startingIndex = endingIndex
endingIndex = endingIndex + trainedDataDatapoints[setClassIndex]
setClassScores[setClasses[setClassIndex]] = score[:, startingIndex:endingIndex-1]
pca_params ={}
pca_params['coeff'] = coeff
pca_params['latent'] = latent
pca_params['dims'] = self.dims
pca_params['window'] = self.window
io.saveYaml('pca_params.yaml', pca_params)
# ======== find average of training sets ========
# find peaks in left score
setClassAverages = {}
for setClassIndex in range(0, len(setClasses)):
setClassAverage = []
setClassScore = setClassScores[setClasses[setClassIndex]]
for peakNumber in range(0, self.numberOfNods):
(d, dp) = setClassScore.shape
peakLocation = np.argmax(np.absolute(setClassScore))%dp
columnsInWindow = range(peakLocation-self.window/2, peakLocation+self.window/2)
setClassAverage.append(setClassScore[:, columnsInWindow])
setClassScore = np.delete(setClassScore, columnsInWindow, 1)
# compute average
setClassAverages[setClasses[setClassIndex]] = np.mean(np.array(setClassAverage), axis=0)
# normalize
for setClass in setClassAverages:
a = setClassAverages[setClass]
row_sums = np.absolute(a).sum(axis=1)
setClassAverages[setClass] = a / row_sums[:, np.newaxis]
# save scores
io.saveYaml('pca_scores.yaml', setClassAverages)
# plot all scores
print "plotting"
for setClass in setClassAverages:
fig = plt.figure()
ax = fig.add_subplot(111)
ax.plot(np.transpose(setClassAverages[setClass]))
filename = 'graph_' + str(setClass) + '.png'
plt.savefig(filename)
os.system("eog " + filename)
for s in range(0, samplesRight):
sample = pcaInputRight[:, s].tolist()
concatenatedPcaInput.append(sample)
pcaInput = np.transpose(np.array(concatenatedPcaInput))
# ======== do PCA ========
coeff, score, latent = pca.princomp(pcaInput.T, dims)
print np.mean(latent)
leftScore = score[:, 0:samplesLeft]
rightScore = score[:, samplesLeft:samplesLeft+samplesRight]
io.saveYaml('pcaTrainingCoeff.yaml', coeff)
# ======== find average of training sets ========
# find peaks in left score
leftScoreAverage = []
for peakNumber in range(0, numberOfNods):
peakLocation = np.argmax(np.absolute(leftScore))%samplesLeft
columnsToDelete = range(peakLocation-window/2, peakLocation+window/2)
leftScoreAverage.append(leftScore[:, columnsToDelete])
leftScore = np.delete(leftScore, columnsToDelete, 1)
leftScore = np.mean(np.array(leftScoreAverage), axis=0)
# save left score
io.saveYaml('trained_left_nod_score.yaml', leftScore)