def learn_metal_cdtw(parentLabels, parent, childLabels, child,
symmetric, squared, parameters=None):
"""learn optimal parameters for gem on metal data"""
if parameters == None:
parameters = np.array(range(21))*0.01
errors = []
for relative_window in parameters:
window = int(np.round(len(parent[0])*relative_window))
func = lambda query, subject: dist.cdtw(query, subject, window, squared)
e, l = c.obtain_1NN_error(parentLabels, parent,
childLabels, child, func)
errors.append((e, (relative_window, squared)))
print errors[-1]
errs = list(sorted(errors, key=lambda x: x[0][0]))
# pick only the best and take parameter in the "middle"
best = filter(lambda (x, y): x[0]==errs[0][0][0], errs)
best.sort(key=lambda (x, y): y[0])
best = best[len(best)/2]
return best, errs
def learn_metal_cdtw(parentLabels,
parent,
childLabels,
child,
symmetric,
squared,
parameters=None):
"""learn optimal parameters for gem on metal data"""
if parameters == None:
parameters = np.array(range(21)) * 0.01
errors = []
for relative_window in parameters:
window = int(np.round(len(parent[0]) * relative_window))
func = lambda query, subject: dist.cdtw(query, subject, window, squared
)
e, l = c.obtain_1NN_error(parentLabels, parent, childLabels, child,
func)
errors.append((e, (relative_window, squared)))
print errors[-1]
errs = list(sorted(errors, key=lambda x: x[0][0]))
# pick only the best and take parameter in the "middle"
best = filter(lambda (x, y): x[0] == errs[0][0][0], errs)
best.sort(key=lambda (x, y): y[0])
best = best[len(best) / 2]
return best, errs
def loocv_cdtw(loocvLabels, loocvSet, relative_window, squared):
"""calculate LOOCV error rate for given window on a given set"""
error, window = 0, int(np.round(relative_window*len(loocvSet[0])))
# 1NN with Leave One Out Cross Validation
for index, (queryLabel, query) in enumerate(zip(loocvLabels, loocvSet)):
search = np.vstack((loocvSet[:index], loocvSet[index+1:]))
labels = np.hstack((loocvLabels[:index], loocvLabels[index+1:]))
vals =[dist.cdtw(query, subject, window, squared) for subject in search]
best = np.array(vals).argmin()
if queryLabel != labels[best]:
error +=1
return error, len(loocvSet), float(error)/len(loocvSet)
def loocv_cdtw(loocvLabels, loocvSet, relative_window, squared):
"""calculate LOOCV error rate for given window on a given set"""
error, window = 0, int(np.round(relative_window * len(loocvSet[0])))
# 1NN with Leave One Out Cross Validation
for index, (queryLabel, query) in enumerate(zip(loocvLabels, loocvSet)):
search = np.vstack((loocvSet[:index], loocvSet[index + 1:]))
labels = np.hstack((loocvLabels[:index], loocvLabels[index + 1:]))
vals = [
dist.cdtw(query, subject, window, squared) for subject in search
]
best = np.array(vals).argmin()
if queryLabel != labels[best]:
error += 1
return error, len(loocvSet), float(error) / len(loocvSet)
