if __name__ == '__main__':
# global parameters for distance measures (Manhatten/Euclidean, sym. gem)
squared, symmetric = True, True
# read dataset number and split number
try:
datasetN, splitN = int(sys.argv[1]), int(sys.argv[2])
except:
raise Exception \
("python2 1NNResampledTest.py datasetnumber(int) splitnumber(int)")
# read the dataset
(testLabels, testSet), (trainLabels, trainSet) = ucr.read(datasetN)
# merge dataset
labels, items = ucr.merge(testLabels, testSet, trainLabels, trainSet)
# open file for the logging of results
f = open("./results/dn_%s-sn_%s-sq_%s-sy_%s" %
(datasetN, splitN, squared, symmetric), "w")
# if splitN == 0 use UCR-split else use random split
if splitN > 0:
# determine split ratio from UCR canonical split and resample
rho = float(len(trainSet))/(len(trainSet)+len(testSet))
sss = st.StratifiedShuffleSplit(labels, 500,
test_size=rho, random_state=0)
test_index, train_index = list(sss)[splitN-1]
bestlp, bestfulldtw, bestconstdtw, bestgem = None, None, None, None
for line in f:
if "BESTLP=" in line:
bestlp=eval(line.split("=")[1])
if "BESTFULLDTW=" in line:
bestfulldtw=eval(line.split("=")[1])
if "BESTCONSDTW=" in line:
bestconsdtw=eval(line.split("=")[1])
if "BESTGEM=" in line:
bestgem=eval(line.split("=")[1])
gainresult.append((bestconsdtw[2]-bestgem[2]))
cdtwresult.append(bestconsdtw[2])
gemresult.append(bestgem[2])
eucresult.append(bestlp[2])
dtwresult.append(bestfulldtw[2])
if np.mean(gainresult) < 0:
print datasetN, "\t", ucr.datasetName(datasetN), "%1.4f +/- %1.4f" % (np.mean(gainresult), np.std(gainresult)), \
" %1.4f" % (np.mean(gainresult)/np.std(gainresult)),
else:
print datasetN, "\t", ucr.datasetName(datasetN), " %1.4f +/- %1.4f" % (np.mean(gainresult),np.std(gainresult)), \
" %1.4f" % (np.mean(gainresult)/np.std(gainresult)),
print " %1.4f +/- %1.4f" % (np.mean(cdtwresult), np.std(cdtwresult)),
print " %1.4f +/- %1.4f" % (np.mean(gemresult), np.std(gemresult)),
print " %1.4f +/- %1.4f" % (np.mean(eucresult), np.std(eucresult)),
print " %1.4f +/- %1.4f" % (np.mean(dtwresult), np.std(dtwresult))
import utils.kNNClassifier as cl
if __name__ == '__main__':
# global parameters for distance measures (Manhatten/Euclidean, sym. gem)
squared, symmetric = True, True
# read dataset number and split number
try:
datasetN, splitN = int(sys.argv[1]), int(sys.argv[2])
except:
raise Exception \
("python2 1NNResampledTest.py datasetnumber(int) splitnumber(int)")
# read the dataset
(testLabels, testSet), (trainLabels, trainSet) = ucr.read(datasetN)
# merge dataset
labels, items = ucr.merge(testLabels, testSet, trainLabels, trainSet)
# open file for the logging of results
f = open(
"./results/dn_%s-sn_%s-sq_%s-sy_%s" %
(datasetN, splitN, squared, symmetric), "w")
# if splitN == 0 use UCR-split else use random split
if splitN > 0:
# determine split ratio from UCR canonical split and resample
rho = float(len(trainSet)) / (len(trainSet) + len(testSet))
sss = st.StratifiedShuffleSplit(labels,
500,
bestlp, bestfulldtw, bestconstdtw, bestgem = None, None, None, None
for line in f:
if "BESTLP=" in line:
bestlp = eval(line.split("=")[1])
if "BESTFULLDTW=" in line:
bestfulldtw = eval(line.split("=")[1])
if "BESTCONSDTW=" in line:
bestconsdtw = eval(line.split("=")[1])
if "BESTGEM=" in line:
bestgem = eval(line.split("=")[1])
gainresult.append((bestconsdtw[2] - bestgem[2]))
cdtwresult.append(bestconsdtw[2])
gemresult.append(bestgem[2])
eucresult.append(bestlp[2])
dtwresult.append(bestfulldtw[2])
if np.mean(gainresult) < 0:
print datasetN, "\t", ucr.datasetName(datasetN), "%1.4f +/- %1.4f" % (np.mean(gainresult), np.std(gainresult)), \
" %1.4f" % (np.mean(gainresult)/np.std(gainresult)),
else:
print datasetN, "\t", ucr.datasetName(datasetN), " %1.4f +/- %1.4f" % (np.mean(gainresult),np.std(gainresult)), \
" %1.4f" % (np.mean(gainresult)/np.std(gainresult)),
print " %1.4f +/- %1.4f" % (np.mean(cdtwresult), np.std(cdtwresult)),
print " %1.4f +/- %1.4f" % (np.mean(gemresult), np.std(gemresult)),
print " %1.4f +/- %1.4f" % (np.mean(eucresult), np.std(eucresult)),
print " %1.4f +/- %1.4f" % (np.mean(dtwresult), np.std(dtwresult))
