def_c_value = cfg.getfloat('svm', 'def_c_value')
# Opens a text file to note test results
text_file = open("output2.txt", "a")
text_file.write("Method used: " + params['method'] + "\n")
text_file.write("-------------------------------------\n")
# If c_iteration is True, iterates from 0.0001 to 10000 for c values.
# Else, uses default c value stated in config.ini file
if c_iteration is True:
for i in range(-4, 5):
BG_img = compute_BG_Image(params['folder'], train_indexes)
trf, trl, trfc = train(train_indexes, BG_img, params)
svm_param = ' -s 0 -t 0 -c ' + str(pow(10, i))
print "c value is: " + str(pow(10, i))
svm = train_svm(trf, trl, svm_param)
trf2, trl2 = bootstrap(bootstrap_indexes, BG_img, params, trf, trl,
trfc, svm)
svm2 = train_svm(trf2, trl2, svm_param)
svm_AP, svm_PR, svm_RC = sw_search(test_indexes, BG_img, params, svm2)
ap_mean = np.mean(svm_AP)
text_file.write("precision for c value " + str(pow(10, i)) + " is: " +
str(ap_mean) + "\n")
else:
BG_img = compute_BG_Image(params['folder'], train_indexes)
trf, trl, trfc = train(train_indexes, BG_img, params)
svm_param = ' -s 0 -t 0 -c ' + str(def_c_value)
svm = train_svm(trf, trl, svm_param)
# - Starts a new job
# - Starts training positive and negative features
#################################################################
if mode == 'train':
details = raw_input("Please enter details about new job: ")
BG_img = compute_BG_Image(params['folder'], train_indexes)
job = push_new_job(method, BG_img, details)
print job['job_id'] + ' job is created at ' + job['timestamp']
print 'Feature extraction method used: ' + job['method']
trf, trl, trfc = train(train_indexes, BG_img, params)
push_train_features(job['job_id'], trf, trl)
svm = train_svm(trf, trl,' -s 0 -t 0 -c 100')
trf, trl = bootstrap(bootstrap_indexes, BG_img, params, trf, trl, trfc, svm)
push_bootstrap_features(job['job_id'], trf, trl)
svm2 = train_svm(trf, trl, '-s 0 -t 0 -c 100')
###################################################################
# If bootstrap mode is selected:
# - Lists last x jobs for user to choose which one to load
# - Retrieves selected training features of selected job
# - Start bootstrapping
###################################################################
elif mode == 'bootstrap':
old_job = find_old_job()
# - Starts a new job
# - Starts training positive and negative features
#################################################################
if mode == 'train':
details = raw_input("Please enter details about new job: ")
BG_img = compute_BG_Image(params['folder'], train_indexes)
job = push_new_job(method, BG_img, details)
print job['job_id'] + ' job is created at ' + job['timestamp']
print 'Feature extraction method used: ' + job['method']
trf, trl, trfc = train(train_indexes, BG_img, params)
push_train_features(job['job_id'], trf, trl)
svm = train_svm(trf, trl, ' -s 0 -t 0 -c 100')
trf, trl = bootstrap(bootstrap_indexes, BG_img, params, trf, trl, trfc,
svm)
push_bootstrap_features(job['job_id'], trf, trl)
svm2 = train_svm(trf, trl, '-s 0 -t 0 -c 100')
###################################################################
# If bootstrap mode is selected:
# - Lists last x jobs for user to choose which one to load
# - Retrieves selected training features of selected job
# - Start bootstrapping
###################################################################
elif mode == 'bootstrap':
old_job = find_old_job()
# Opens a text file to note test results
text_file = open("output2.txt", "a")
text_file.write("Method used: " + params['method'] + "\n")
text_file.write("-------------------------------------\n")
# If c_iteration is True, iterates from 0.0001 to 10000 for c values.
# Else, uses default c value stated in config.ini file
if c_iteration is True:
for i in range(-4, 5):
BG_img = compute_BG_Image(params['folder'], train_indexes)
trf, trl, trfc = train(train_indexes, BG_img, params)
svm_param = ' -s 0 -t 0 -c ' + str(pow(10, i))
print "c value is: " + str(pow(10, i))
svm = train_svm(trf, trl, svm_param)
trf2, trl2 = bootstrap(bootstrap_indexes, BG_img, params, trf, trl, trfc, svm)
svm2 = train_svm(trf2, trl2, svm_param)
svm_AP, svm_PR, svm_RC = sw_search(test_indexes, BG_img, params, svm2)
ap_mean = np.mean(svm_AP)
text_file.write("precision for c value " + str(pow(10, i)) + " is: " + str(ap_mean) + "\n")
else:
BG_img = compute_BG_Image(params['folder'], train_indexes)
trf, trl, trfc = train(train_indexes, BG_img, params)
svm_param = ' -s 0 -t 0 -c ' + str(def_c_value)
svm = train_svm(trf, trl, svm_param)
trf2, trl2 = bootstrap(bootstrap_indexes, BG_img, params, trf, trl, trfc, svm)