def test_calculate_self_confusion_vector(self):
n_items = 6
report, image_sets = rate_fp.get_docs(n_items=n_items)
report = rate_fp.calculate_self_confusion_vector(
image_sets,
fingerprint_function=fingerprint_core.fp,
weights=np.ones(constants.fingerprint_length),
distance_function=NNSearch.distance_1_k,
distance_power=0.5,
report=report,
parallelize=False)
avg = report['average_weighted']
print('self item avg without parallelization:' + str(avg))
self.assertTrue(avg > 0 and avg < 100) # this 100 is arbirary...
report = rate_fp.calculate_self_confusion_vector(
image_sets,
fingerprint_function=fingerprint_core.fp,
weights=np.ones(constants.fingerprint_length),
distance_function=NNSearch.distance_1_k,
distance_power=0.5,
report=report,
parallelize=True)
avg2 = report['average_weighted']
print('self item avg with parallelization:' + str(avg2))
self.assertTrue(avg == avg2) # this 100 is arbirary...
def test_get_docs(self):
n_items = 5
report, docs = rate_fp.get_docs(n_items=n_items)
l = len(docs)
print('n:' + str(l) + ' docs:' + str(docs))
print('report:' + str(report))
self.assertTrue(l == n_items)
def optimize_weights(weights=np.ones(fingerprint_length),k=0.5):
'''
optimizes weights given everything else constant (fingerprint function, distance function, distance power k)
:param k: distance power - k=0.5 is for euclidean distance
:return:optimal weights vector (or at least best found so far)
'''
print('fp length:'+str(fingerprint_length))
print('weights:'+str(weights))
# x_min = scipy.optimize.minimize(f,initial_weights,args=(k),tol=0.1)
f = rate_wrapper
init=weights
# TO DO CONSTRAINED MINIMIZATION USE COBYLA or SQLSQP - see docs - currentyly constraining 'by hand'
# x_min = scipy.optimize.minimize(f,init,args=(k,),tol=0.1) #in case u need only one optional argument this is how to do it
pr = cProfile.Profile()
pr.enable()
self_report, image_sets = rate_fp.get_docs(n_docs)
comparisons_to_make = rate_fp.make_cross_comparison_sets(image_sets)
print('k:' + str(k) + 'len imsets:' + str(len(image_sets)) + 'selfrep:' + str(self_report) )
name = 'report.' + datetime.datetime.now().strftime("%Y-%m-%d.%H%M")
outfilename = os.path.join('fp_ratings', name)
print(name)
x_min = scipy.optimize.minimize(f, init, args=(k, image_sets, self_report, comparisons_to_make, outfilename),
tol=0.05,
options={'maxiter': 50, 'disp': True})
pr.disable()
s = StringIO.StringIO()
sortby = 'cumulative'
ps = pstats.Stats(pr, stream=s).sort_stats(sortby)
ps.print_stats()
print(s.getvalue())
# x_min = scipy.optimize.fmin(f,[2,3])
print('output of optimize:'+str(x_min))
print('xvals:'+str(x_min.x))
print('f('+str(x_min.x)+')='+str(f(x_min.x)))
def test_make_cross_comparison_sets(self):
n_items = 4
report, image_sets = rate_fp.get_docs(n_items=n_items)
sets = rate_fp.make_cross_comparison_sets(image_sets)
print(sets)
self.assertTrue(len(sets) == n_items)