def train_all_models():
with open('accuracy.csv', 'w') as f:
f.write('category,tp,fp,tn,fn,precision,recall,fscore')
f.write('\n')
for c in categories:
generate(c)
train_for_category(c, 'KNN')
def run_T_detection_trials(trials=100, size=1000, cprivacy=0.1, nprivacy=1, distinct=10, cskew=0.33, nskew=0.33):
failure = 0
for t in range(0, trials):
dataset = generate(S=size, p=cprivacy, b=nprivacy, N=distinct, z1=cskew, z2=nskew)
res = distinct_count(dataset)
failure = failure + (res[0] == res[2])
return (failure + 0.0) / trials
def train_with_optimized_feature_size(category, classifier):
clfs_with_score = {}
for i in range(10, 14):
with open('num_of_top_words_as_feature.pickle', 'w+b') as f:
num_of_top_words_as_feature = i
pickle.dump(num_of_top_words_as_feature, f)
generate(category)
f1_score, cm, clf = train_for_category(category, classifier)
clfs_with_score[i] = f1_score
max_score = 0
best_feature_size = 10
clf = None
for i in clfs_with_score:
if max_score < clfs_with_score[i]:
print i
best_feature_size = i
max_score = clfs_with_score[i]
print category + " " + str(max_score) + " " + str(best_feature_size)
with open('num_of_top_words_as_feature.pickle', 'w+b') as f:
pickle.dump(best_feature_size, f)
#trained again so that all csv files are updated for correct feature size
generate(category)
f1_score, cm, clf = train_for_category(category, classifier)
print f1_score
with open('model/' + category, 'w+b') as f:
pickle.dump(clf, f)
with open('data/' + category + '/feature_size.pickle', 'w') as f:
pickle.dump(best_feature_size, f)
return clf, max_score
def run_T_trials(
trials=100,
size=1000,
cprivacy=0.1,
nprivacy=1,
distinct=10,
selectivity=0.5,
errorrate=0.21,
cskew=0.33,
nskew=0.33,
):
sumr = []
countr = []
avgr = []
for t in range(0, trials):
# print "Running Trial ", t
dataset = generate(S=size, p=cprivacy, b=nprivacy, N=distinct, z1=cskew, z2=nskew)
truepred = range(
int(round((1 - selectivity) * distinct)), distinct
) # numpy.random.choice(range(0,distinct),size=int(round(selectivity*distinct)),replace=False)
corruptpred = []
if int(round(errorrate * distinct)) >= 1:
corruptpred = numpy.random.choice(range(0, distinct), size=int(round(errorrate * distinct)), replace=False)
corruptpred = list(set(truepred).union(set(corruptpred)))
sum_cresult = sumq(dataset, predicate=truepred, p=cprivacy, b=nprivacy)
count_cresult = countq(dataset, predicate=truepred, p=cprivacy)
avg_cresult = avgq(dataset, predicate=truepred, p=cprivacy, b=nprivacy)
sum_dresult = sumq(dataset, predicate=corruptpred, p=cprivacy, b=nprivacy)
count_dresult = countq(dataset, predicate=corruptpred, p=cprivacy)
avg_dresult = avgq(dataset, predicate=corruptpred, p=cprivacy, b=nprivacy)
if count_cresult[2] > 0:
sumr.append((sum_cresult[0], sum_dresult[1], sum_cresult[2], sum_dresult[2], sum_cresult[1]))
countr.append((count_cresult[0], count_dresult[1], count_cresult[2], count_dresult[2], count_cresult[1]))
avgr.append((avg_cresult[0], avg_dresult[1], avg_cresult[2], avg_dresult[2], count_cresult[1]))
return (sumr, countr, avgr)
batch_size=100,
max_anchor_size=None,
stop_on_first=False,
desired_label=None,
beam_size=4)
print('Anchor: %s' % (' AND '.join(exp.names())))
coverage_tab = None
precision_tab = None
# If the dataset is one of the 3 generated, pick the anchors informations and compute the coverage and precision on test dataset
if 'generate' in dataset_name:
anchors = exp.names()
_, _, pick_anchors_informations = generate(
dataset.test[idx].reshape(1, -1)[0][0],
dataset.test[idx].reshape(1, -1)[0][1],
anchors,
blackbox=c,
X=X,
Y=Y)
coverage_tab = compute_coverage(dataset, exp, anchors,
pick_anchors_informations)
precision_tab = compute_precision(
dataset, dataset.labels_test,
predict_fn(dataset.test[idx].reshape(1, -1))[0], exp,
anchors, pick_anchors_informations)
"""
print('Anchor: %s' % (' AND '.join(exp.names())))
print('Precision: %.2f' % exp.precision())
print('Coverage: %.2f' % exp.coverage())
"""
choice = input("Generate dataset? [y/n]: ")
iterations = 5
mode = CameraPose.NONE
csvfile = './outputs/flowers_dataset.csv'
csvMode = 'w' # 'a' => append / 'w' => write (overwrites existing csv file)
meshes = [(flower_mesh, 'flower'), (stem_mesh, 'stem'), (center_mesh, 'center')]
skip_default_view = False
salt = '001'
offset = 0
skip_sub_mesh = {name: False for (mesh, name) in meshes}
if flower_type in no_stem_flowers:
skip_sub_mesh['stem'] = True
if choice == 'y':
# Remove camera that was used as preview from the scene
scene.remove_node(cam_node)
generate(
scene=scene, renderer=r, flags=flags, alpha=alpha, beta=beta, camera=camera,
iterations=iterations, flower_pose=flower_pose, meshes=meshes,
light_intensitites = varying_light_intensities,
skip_sub_mesh= skip_sub_mesh, at_x=at_x, at_y=at_y, at_z=at_z,
skip_default_view=skip_default_view, offset=offset,
distance=distance, mode=mode, csvfile=csvfile, csvMode=csvMode,
flower_name=file_name, salt=salt
)
print("Finished")
