def classification_preprocess_all_datasets():
"""
Preprocesses all datasets to be ready for classification task.
This will include stemming, word correction, lower-casing, hashtag removal, special char removal.
"""
for i in range(0,len(utils.annotated_datasets)):
tweetlines = utils.get_dataset(utils.annotated_datasets[i])
tweets = []
for line in tweetlines:
if len(line)>1:
tweets.append(tweet.to_tweet(line))
# tweets = lower_case(tweets)
tweets = remove_hastags_and_users(tweets)
tweets = count_emoticons(tweets)
tweets = replace_links(tweets)
tweets = remove_specialchars(tweets)
tweets = correct_words(tweets)
tweets = stem(tweets)
tweets = tokenize(tweets)
tweets = pos_tag(tweets)
tweets = count_exclamations(tweets)
analyzer = Analyzer(utils.annotated_datasets[i], tweets)
stats = analyzer.analyze()
print stats
#store tweets in pickles...
print "Storing pickles..."
utils.store_pickles(tweets, utils.annotated_datasets[i][24:len(utils.annotated_datasets[i])-4])
def preprocess_temporal_dataset():
tweetlines = utils.get_dataset(utils.complete_datasets[3])
tweets = []
for line in tweetlines:
if len(line) > 1:
tweets.append(tweet.to_tweet(line))
tweets = preprocessing.preprocess_tweets(tweets)
sentiments = lexicon.perform_google_sentiment_lexicon_lookup(tweets)
pickle.dump(sentiments, open("temporal_sentiments", "wb"))
pickle.dump(tweets, open("temporal_tweets2", "wb"))
def initial_preprocess_all_datasets():
"""
Runs first preprocessing iteration on all datasets.
This is the preprocessing routine performed initially on the datasets before annotation.
This routine includes duplicate removal
"""
for i in range(0,len(utils.datasets)):
#Fetch from dataset
tweets = []
tweetlines = utils.get_dataset(utils.complete_datasets[i])
for tweetline in tweetlines:
tweets.append(tweet.to_tweet(tweetline))
#Perform preprocessing
tweets = remove_duplicates_and_retweets(tweets)
#Store back to dataset
tweetlines = []
for t in tweets:
tweetlines.append(t.to_tsv())
utils.store_dataset(tweetlines, utils.datasets[i])
def test_supermodel(datasets, args, n_folds=5, times=20, verbose=False):
for item in datasets:
print(type(item))
if isinstance(item, tuple):
X, y = item[0], item[1]
else:
X, y, _ = utils.get_dataset(item)
print("TESTING %s" % item.upper())
print("Dataset shape: %s, %s" % (X.shape, y.shape))
args["dataset_name"] = item
super_res, classifiers_res = scores_stats(X, y, cv_type="stratified", n_folds=5, times=times, verbose=False, args=args)
print("Classifiers results:")
max_mean = 0
max_std = 0
for idx, val in enumerate(classifiers_res[0]):
mean, std = (classifiers_res[0][idx], classifiers_res[1][idx])
if mean > max_mean:
max_mean = mean
max_std = std
print("Mean = %.4f, std = %.4f" % (mean, std))
print("Max classifier mean:\nMean = %.4f, std = %.4f" % (max_mean, max_std))
print("Superclassifier score")
print("Mean = %.4f, std = %.4f" % (super_res[0], super_res[1]))
np.savez(os.path.join(save_dir, video_name + '.npz'),
avg_score=avg_score.mean(0).cpu().numpy(),
weight=weight.mean(0).cpu().numpy()
if weight is not None else None,
global_score=global_score.mean(0).cpu().numpy(),
branch_scores=branch_scores)
if args.include_train:
train_dataset_dict = get_dataset(
dataset_name=dataset_name,
subset=args.train_subset_name,
file_paths=file_paths,
sample_rate=sample_rate,
base_sample_rate=base_sample_rate,
action_class_num=action_class_num,
modality='flow',
feature_type=feature_type,
feature_oversample=feature_oversample,
temporal_aug=False,
)
train_detect_dataset = SingleVideoDataset(
train_dataset_dict, single_label=False,
random_select=False) # SIngle label false!!!
train_detect_loader = torch.utils.data.DataLoader(train_detect_dataset,
batch_size=1,
pin_memory=True,
shuffle=False)
def get_data_loaders(args, tokenizer):
""" Prepare the dataset for training and evaluation """
personachat = get_dataset(tokenizer, args.dataset_path, args.dataset_cache)
logger.info("Build inputs and labels")
datasets = {"train": defaultdict(list), "valid": defaultdict(list)}
for dataset_name, dataset in personachat.items():
num_candidates = len(dataset[0]["utterances"][0]["candidates"])
if args.num_candidates > 0 and dataset_name == 'train':
num_candidates = min(args.num_candidates, num_candidates)
for dialog in dataset:
persona = dialog["personality"].copy()
for _ in range(
max(1, min(args.personality_permutations, len(persona)))):
for utterance in dialog["utterances"]:
history = utterance["history"][-(2 * args.max_history +
1):]
for j, candidate in enumerate(
utterance["candidates"][-num_candidates:]):
lm_labels = bool(j == num_candidates - 1)
instance = build_input_from_segments(
persona, history, candidate, tokenizer, lm_labels)
for input_name, input_array in instance.items():
datasets[dataset_name][input_name].append(
input_array)
datasets[dataset_name]["mc_labels"].append(num_candidates -
1)
datasets[dataset_name]["n_candidates"] = num_candidates
if len(persona) > 1:
persona = [persona[-1]
] + persona[:-1] # permuted personalities
logger.info("Pad inputs and convert to Tensor")
tensor_datasets = {"train": [], "valid": []}
for dataset_name, dataset in datasets.items():
dataset = pad_dataset(dataset,
padding=tokenizer.convert_tokens_to_ids(
SPECIAL_TOKENS[-1]))
for input_name in MODEL_INPUTS:
tensor = torch.tensor(dataset[input_name])
if input_name != "mc_labels":
tensor = tensor.view((-1,
datasets[dataset_name]["n_candidates"]) +
tensor.shape[1:])
tensor_datasets[dataset_name].append(tensor)
logger.info("Build train and validation dataloaders")
train_dataset, valid_dataset = TensorDataset(
*tensor_datasets["train"]), TensorDataset(*tensor_datasets["valid"])
train_sampler = torch.utils.data.distributed.DistributedSampler(
train_dataset) if args.distributed else None
valid_sampler = torch.utils.data.distributed.DistributedSampler(
valid_dataset) if args.distributed else None
train_loader = DataLoader(train_dataset,
sampler=train_sampler,
batch_size=args.train_batch_size,
shuffle=(not args.distributed))
valid_loader = DataLoader(valid_dataset,
sampler=valid_sampler,
batch_size=args.valid_batch_size,
shuffle=False)
logger.info("Train dataset (Batch, Candidates, Seq length): {}".format(
train_dataset.tensors[0].shape))
logger.info("Valid dataset (Batch, Candidates, Seq length): {}".format(
valid_dataset.tensors[0].shape))
return train_loader, valid_loader, train_sampler, valid_sampler
special_chars_removal = '(<|>|{|}|[|]|-|_|*|")'
replacement_chars = {u"&": u"og",
u"6amp;": u"og",
u"+": u"og"}
if __name__ == '__main__':
#Testing
# tweets = [Tweet("13:37", "johnarne", "Jeg () haaater drittt!!!? :( #justinbieber"), Tweet("13:37", "johnarne", "Jeg eeelsker @erna_solberg http://www.erna.no :) #love #jernerna" )]
# for tweet in tweets:
# tweet.set_sentiment("negative")
# print tweet
tweetlines = utils.get_dataset("test_annotated_data/erna_dataset.tsv")
tweets = []
for line in tweetlines:
if len(line)>1:
tweets.append(tweet.to_tweet(line))
# tweets = lower_case(tweets)
tweets = remove_hastags_and_users(tweets)
tweets = count_emoticons(tweets)
tweets = replace_links(tweets)
tweets = remove_specialchars(tweets)
for tweet in tweets:
print tweet
tweets = correct_words(tweets)
tweets = stem(tweets)
tmp_X_test = utils.remove_feature(tmp_X_test, 11)
tmp_X_test = utils.remove_feature(tmp_X_test, 10)
tmp_X_test = utils.remove_feature(tmp_X_test, 9)
tmp_X_test = utils.remove_feature(tmp_X_test, 8)
tmp_X_test = utils.remove_feature(tmp_X_test, 7)
tmp_X_test = utils.remove_feature(tmp_X_test, 6)
# tmp_X_test = utils.remove_feature(tmp_X_test, 5)
# tmp_X_test = utils.remove_feature(tmp_X_test, 4)
# tmp_X_test = utils.remove_feature(tmp_X_test, 3)
# tmp_X_test = utils.remove_feature(tmp_X_test, 2)
# tmp_X_test = utils.remove_feature(tmp_X_test, 1)
# tmp_X_test = utils.remove_feature(tmp_X_test, 0)
return tmp_X_train, tmp_X_test
X, Y = utils.get_dataset('dataset_threshold_100.txt')
# X = preprocessing.normalize(X)
Y = utils.convert_Y_to_class_numbers(Y)
# X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=0.2)
X_train, X_test, Y_train, Y_test = train_test_split(X,
Y,
test_size=0.2,
random_state=50)
# X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=0.2, random_state=40)
# X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=0.2, random_state=30)
# X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=0.2, random_state=20)
# X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=0.2, random_state=10)
# X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=0.2, random_state=0)
# X_train, X_test = prepare_dataset(X_train, X_test)
# plot_accuracy(X_train, Y_train, X_test, Y_test, KNeighborsClassifier(n_neighbors=5), 'KNN')
def associate(real_world: bool, audio: bool):
project_dir = os.getcwd()
cfg = utils.get_config(project_dir)
bio_type = 'audio' if audio else 'video'
data_path = os.path.join(
project_dir, 'data',
cfg['base_conf']['dataset'][utils.get_dataset(real_world, bio_type)])
bio_data_path = os.path.join(data_path, cfg['base_conf']['biometric_data'])
wifi_data_path = os.path.join(data_path, cfg['base_conf']['wifi_data'])
wifi_thres = cfg['parameters']['wifi_threshold']
meeting_thres = cfg['parameters']['meeting_threshold']
cdist_metric = cfg['parameters']['cdist_metric']
victims_thres = cfg['parameters']['estimated_victims']
omega = cfg['parameters']['omega']
# meeting_npy_paths, the npy path that contains feature vectors, after running feature_extraction
# {'meeting_15': 'CrossLeak/data/audio_50vs20_100/bio_data/meeting_15/vecmeeting_15.npy'}}
meeting_npy_paths = utils.get_meeting_and_path(bio_data_path, r'.+\.npy$')
# {'meeting_51': 'CrossLeak/data/audio_50vs20_100/bio_data/meeting_51/segsmeeting_51.pk'}
bio_path = utils.get_meeting_and_path(bio_data_path, r'.+segs.+\.pk$')
assert len(meeting_npy_paths.keys()) == len(bio_path.keys())
# get participants' names, given the corresponding WiFi sniffing files (wifi_thres is the cutoff rss threshold)
meeting_people_name = utils.get_meeting_people_name(
wifi_data_path, real_world, r'.+\.pk$', wifi_thres)
# load and map poi name and mac address, respectively
poi_name_mac = {}
meeting_poi_name = collections.defaultdict(list)
if real_world:
for mac, poi in cfg['mac_name']:
if poi not in poi_name_mac:
poi_name_mac[poi] = mac
else:
logging.warning(
'duplicate mac address of {} with {} and {}'.format(
poi, poi_name_mac[poi], mac))
else:
poi = pickle.load(open(os.path.join(bio_data_path, 'POIs.pk'), 'rb'))
non_poi = pickle.load(
open(os.path.join(bio_data_path, 'nonPOIs.pk'), 'rb'))
for peo in poi:
poi_name_mac[peo] = peo
for meeting, peos in meeting_people_name.items():
for peo in peos:
if peo in poi:
meeting_poi_name[meeting].append(peo)
elif peo not in non_poi:
logging.warning('{} not in poi nor non_poi'.format(peo))
# remove filter people
excluded_peo = cfg['filter_people'][bio_type]
for meeting, people_names in meeting_poi_name.items():
meeting_poi_name[meeting] = [
poi for poi in people_names if poi not in excluded_peo
]
# flatten to get unique poi name
poi_people = sorted({
name
for people_names in meeting_poi_name.values() for name in people_names
})
logging.info('poi people: {}'.format(poi_people))
meeting_name = sorted(meeting_poi_name.keys())
meeting_num = len(meeting_name)
if meeting_thres == -1:
meeting_thres = meeting_num
# get meeting index by meeting name
meeting_index = dict(zip(meeting_name, list(range(meeting_num))))
poi_num = len(poi_people)
context_infor = np.zeros([meeting_num, poi_num]).astype(np.float64)
for meeting, people_names in meeting_poi_name.items():
for poi in people_names:
context_infor[meeting_index[meeting], poi_people.index(poi)] = 1
bio_info = np.empty([0, cfg['pre_process'][bio_type]['dimension']])
bio_paths = []
for meeting in bio_path.keys():
with open(bio_path[meeting], 'rb') as f:
relative_bio_paths = pickle.load(f)
absolute_bio_paths = [
os.path.join(bio_data_path, p) for p in relative_bio_paths
]
bio_paths.extend(absolute_bio_paths)
# iou vec is the face / voice features via feature extractor
iou_vec = np.load(meeting_npy_paths[meeting])
try:
bio_info = np.vstack((bio_info, iou_vec))
except:
logging.error('error in numpy vstack: {}'.format(meeting))
# construct mac attendance vector
real_mac_attendance = defaultdict(
lambda: [0] * min(meeting_num, meeting_thres))
if not real_world:
# simply use people name to represent MAC address
meeting_people_mac = utils.get_meeting_people_name(
wifi_data_path, real_world, r'.+\.pk$', wifi_thres)
for meeting, macs in meeting_people_mac.items():
for mac in macs:
real_mac_attendance[mac][meeting_index[meeting]] = 1
# use mac_index to get the index of mac address in mac_attendance
mac_index = {}
poi_mac_attendance = []
cnt = 0
for poi in poi_people:
if poi in poi_name_mac and poi_name_mac[poi] in real_mac_attendance:
# divide real_mac_attendance into two sequences instead of random: first with poi and following non_poi.
poi_mac_attendance.append(real_mac_attendance[poi_name_mac[poi]])
mac_index[cnt] = poi_name_mac[poi]
cnt += 1
real_mac_attendance.pop(poi_name_mac[poi])
else:
# check if all poi has the name-mac mapping and poi attend at least one meeting
logging.error(
'poi {} do not have mac information or have not attended at least one meeting'
.format(poi))
assert len(real_mac_attendance) == 0
# add the non poi part of mac_attendance
non_poi_mac_attendance = []
if real_world:
for mac, attendance in real_mac_attendance.items():
mac_index[cnt] = mac
cnt += 1
non_poi_mac_attendance.append(attendance)
else:
non_poi = pickle.load(
open(os.path.join(bio_data_path, 'nonPOIs.pk'), 'rb'))
non_poi_mac_attendance = [[
0 for _ in range(min(meeting_num, meeting_thres))
] for _ in range(len(non_poi))]
for oos_idx in range(len(non_poi)):
random_rssi = np.random.normal(-60, 80,
min(meeting_num, meeting_thres))
for meeting in range(min(meeting_num, meeting_thres)):
if random_rssi[meeting] >= wifi_thres:
non_poi_mac_attendance[oos_idx][meeting] = 1
mac_attendance = np.concatenate(
(poi_mac_attendance, non_poi_mac_attendance))
if real_world:
# remove always-on mac address, detected over 90% of all meetings, which may be router or something.
always_on_index = np.where(
mac_attendance.sum(axis=1) > 0.9 * mac_attendance.shape[1])[0]
mac_attendance = np.delete(mac_attendance, always_on_index, axis=0)
logging.info(
'always on mac address no: {} of all meeting no: {}'.format(
len(always_on_index), mac_attendance.shape[1]))
logging.info('mac attendance len: {} with poi no: {}'.format(
len(mac_attendance), len(poi_mac_attendance)))
# concatenate features and attendance vector (ctx information)
assert len(bio_paths) == len(bio_info)
logging.info('bio_paths and bio_info len: {}'.format(len(bio_paths)))
# get bio_features in valid meeting
bio_features = []
for bio_feature, path in zip(bio_info, bio_paths):
parent = utils.get_parent_folder_name(path, 3)
c = parent.split('_')
if meeting_index['%s_%s' % (c[0], c[1])] < meeting_thres:
bio_features.append(bio_feature)
# event vector of MAC addresses
wifi_people_in_meetings = np.zeros(
[poi_num, min(meeting_num, meeting_thres)])
for i in range(poi_num):
for name in meeting_name:
if meeting_index[name] < meeting_thres:
wifi_people_in_meetings[i,
meeting_index[name]] = context_infor[
meeting_index[name], i]
# for i in range(poi_num):
# for name in meeting_name:
# if poi_people[i] in meeting_poi_name[name]:
# if meeting_index[name] < meeting_thres:
# wifi_people_in_meetings[i, meeting_index[name]] = 1
scan(bio_features, bio_paths, bio_info, real_world, cdist_metric,
victims_thres, omega, meeting_num, meeting_thres, meeting_index,
wifi_people_in_meetings, poi_people, project_dir, bio_data_path,
audio)
flip = args.dataset == "subj"
keywords = get_extended_keyword(keywords,
vocab=None,
weight=args.weight,
n=args.extension)
if pseudo:
if args.extension > 0:
print("Extended Keywords :", keywords)
else:
print("Original Keywords :", keywords)
else:
print("ORACLE MODE")
x_train, x_test, y_test, lp, ln, sc = get_dataset(args.dataset,
key=keywords,
threshold=args.threshold,
pseudo=pseudo,
flip=flip)
# check pseudo-labeling method
tp = len(np.where(lp == 1)[0])
fp = len(np.where(lp == -1)[0])
tn = len(np.where(ln == -1)[0])
fn = len(np.where(ln == 1)[0])
print("--------------------------------------------")
print("Result of Pseudo-labeling Algorithm 1")
print("Pseudo-positive (true-p, false-p) = (%d, %d)" % (tp, fp))
print("Pseudo-negative (false-n, true-n) = (%d, %d)" % (fn, tn))
print("--------------------------------------------")
pi = float(tp / len(lp))
pi_p = float(fn / len(ln))
print(
import pandas as pd
import matplotlib.pyplot as plt
import pickle
parser = argparse.ArgumentParser()
parser.add_argument("--weight-path", required=True, type=str)
parser.add_argument("--model-name", required=True, type=str, help=['preact50'])
parser.add_argument("--data-path",default="data")
parser.add_argument("--dataset", default="sim_real",type=str)
parser.add_argument("--size", default=224, type=int)
parser.add_argument("--batch-size",default=16, type=int)
parser.add_argument("--eval-batch-size", default=64, type=int)
args = parser.parse_args()
train_ds, test_ds = get_dataset(args)
model = get_model(args)
if __name__=='__main__':
model.build((2, 224,224,3)) # Build
model.load_weights(args.weight_path) # Load
# Compile
model.compile(loss='sparse_categorical_crossentropy', optimizer='adam', metrics=['acc'])
# import IPython; IPython.embed();exit(1)
labels, preds = [], []
for img,label in test_ds:
labels += list(label.numpy())
preds += list(model.predict(img).argmax(1))
cm = tf.math.confusion_matrix(labels, preds)
print(cm)
with open("data/label_to_id_dict.dict", "rb") as f:
import pickle
import numpy as np
from utils import get_dataset
import matplotlib.pyplot as plt
from utils import construct
window = 30
fitted_model = pickle.load(open('HMM_tau_' + str(window), 'rb'))
dataset = get_dataset(window)
p_comps = dataset.pca_results.components_
motifs = fitted_model.means_
all_labels = fitted_model.decode(dataset.all_reduced_segments)[1]
unique_labels, counts = np.unique(all_labels, return_counts=True)
sorted_unique_labels = unique_labels[np.argsort(counts)[::-1]]
""" Plot most common three motifs (judged by marginal prob)"""
motif_fig = plt.figure(figsize=(3, 3))
all_motifs = np.array([])
for i in range(10):
true_ind = int(sorted_unique_labels[i])
motif = construct(true_ind, p_comps, motifs)
plt.plot(motif[1, :], -motif[0, :], label=str(i))
plt.legend()
plt.xticks(fontsize=16)
plt.yticks(fontsize=16)
def run():
parser = ArgumentParser()
parser.add_argument(
"--dataset_path",
type=str,
default="",
help="Path or url of the dataset. If empty download from S3.")
parser.add_argument("--dataset_cache",
type=str,
default='./dataset_cache',
help="Path or url of the dataset cache")
parser.add_argument(
"--model",
type=str,
default="openai-gpt",
help="Model type (openai-gpt or gpt2)",
choices=['openai-gpt',
'gpt2']) # anything besides gpt2 will load openai-gpt
parser.add_argument("--model_checkpoint",
type=str,
default="",
help="Path, url or short name of the model")
parser.add_argument(
"--max_history",
type=int,
default=2,
help="Number of previous utterances to keep in history")
parser.add_argument("--device",
type=str,
default="cuda" if torch.cuda.is_available() else "cpu",
help="Device (cuda or cpu)")
parser.add_argument("--no_sample",
action='store_true',
help="Set to use greedy decoding instead of sampling")
parser.add_argument("--max_length",
type=int,
default=20,
help="Maximum length of the output utterances")
parser.add_argument("--min_length",
type=int,
default=1,
help="Minimum length of the output utterances")
parser.add_argument("--seed", type=int, default=0, help="Seed")
parser.add_argument("--temperature",
type=int,
default=0.7,
help="Sampling softmax temperature")
parser.add_argument(
"--top_k",
type=int,
default=0,
help="Filter top-k tokens before sampling (<=0: no filtering)")
parser.add_argument(
"--top_p",
type=float,
default=0.9,
help="Nucleus filtering (top-p) before sampling (<=0.0: no filtering)")
args = parser.parse_args()
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__file__)
logger.info(pformat(args))
if args.model_checkpoint == "":
if args.model == 'gpt2':
raise ValueError(
"Interacting with GPT2 requires passing a finetuned model_checkpoint"
)
else:
args.model_checkpoint = download_pretrained_model()
if args.seed != 0:
random.seed(args.seed)
torch.random.manual_seed(args.seed)
torch.cuda.manual_seed(args.seed)
logger.info("Get pretrained model and tokenizer")
tokenizer_class, model_class = (
GPT2Tokenizer,
GPT2LMHeadModel) if args.model == 'gpt2' else (OpenAIGPTTokenizer,
OpenAIGPTLMHeadModel)
tokenizer = tokenizer_class.from_pretrained(args.model_checkpoint)
model = model_class.from_pretrained(args.model_checkpoint)
model.to(args.device)
add_special_tokens_(model, tokenizer)
logger.info("Sample a personality")
dataset = get_dataset(tokenizer, args.dataset_path, args.dataset_cache)
personalities = [
dialog["personality"] for dataset in dataset.values()
for dialog in dataset
]
personality = random.choice(personalities)
logger.info("Selected personality: %s",
tokenizer.decode(chain(*personality)))
history = []
engine = pyttsx3.init()
r = sr.Recognizer()
while True:
print("Talk:")
with sr.Microphone() as source:
audio = r.listen(source)
raw_text = r.recognize_google(audio)
print(raw_text)
# raw_text = input(">>> ")
while not raw_text:
print('Prompt should not be empty!')
raw_text = input(">>> ")
history.append(tokenizer.encode(raw_text))
with torch.no_grad():
out_ids = sample_sequence(personality, history, tokenizer, model,
args)
history.append(out_ids)
history = history[-(2 * args.max_history + 1):]
out_text = tokenizer.decode(out_ids, skip_special_tokens=True)
print(out_text)
engine.say(out_text)
engine.runAndWait()
class run:
parser = ArgumentParser()
parser.add_argument(
"--dataset_path",
type=str,
default="",
help="Path or url of the dataset. If empty download from S3.")
parser.add_argument("--dataset_cache",
type=str,
default='./dataset_cache',
help="Path or url of the dataset cache")
parser.add_argument(
"--model",
type=str,
default="openai-gpt",
help="Model type (openai-gpt or gpt2)",
choices=['openai-gpt',
'gpt2']) # anything besides gpt2 will load openai-gpt
parser.add_argument("--model_checkpoint",
type=str,
default="",
help="Path, url or short name of the model")
parser.add_argument(
"--max_history",
type=int,
default=20,
help="Number of previous utterances to keep in history")
parser.add_argument("--device",
type=str,
default="cuda" if torch.cuda.is_available() else "cpu",
help="Device (cuda or cpu)")
parser.add_argument("--no_sample",
action='store_true',
help="Set to use greedy decoding instead of sampling")
parser.add_argument("--max_length",
type=int,
default=20,
help="Maximum length of the output utterances")
parser.add_argument("--min_length",
type=int,
default=1,
help="Minimum length of the output utterances")
parser.add_argument("--seed", type=int, default=0, help="Seed")
parser.add_argument("--temperature",
type=int,
default=0.7,
help="Sampling softmax temperature")
parser.add_argument(
"--top_k",
type=int,
default=0,
help="Filter top-k tokens before sampling (<=0: no filtering)")
parser.add_argument(
"--top_p",
type=float,
default=0.9,
help="Nucleus filtering (top-p) before sampling (<=0.0: no filtering)")
args = parser.parse_args()
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__file__)
logger.info(pformat(args))
if args.model_checkpoint == "":
if args.model == 'gpt2':
raise ValueError(
"Interacting with GPT2 requires passing a finetuned model_checkpoint"
)
else:
args.model_checkpoint = download_pretrained_model()
if args.seed != 0:
random.seed(args.seed)
torch.random.manual_seed(args.seed)
torch.cuda.manual_seed(args.seed)
logger.info("Get pretrained model and tokenizer")
tokenizer_class, model_class = (
GPT2Tokenizer,
GPT2LMHeadModel) if args.model == 'gpt2' else (OpenAIGPTTokenizer,
OpenAIGPTLMHeadModel)
tokenizer = tokenizer_class.from_pretrained(args.model_checkpoint)
model = model_class.from_pretrained(args.model_checkpoint)
model.to(args.device)
add_special_tokens_(model, tokenizer)
logger.info("Sample a personality")
dataset = get_dataset(tokenizer, args.dataset_path, args.dataset_cache)
#personalities = [dialog["personality"] for dataset in dataset.values() for dialog in dataset]
#logger.info("Selected personality: %s", tokenizer.decode(chain(*personality)))
history = []
def process_text(self, raw_text):
#personality = random.choice(self.personalities)
personality = [
'i am a robot.', 'my job is to give or deny permission.',
'i love my job.', 'josh is my favorite person.',
'my name is permissioner-bot.', 'i do not have a gender.'
]
personality = [self.tokenizer.encode(line) for line in personality]
self.history.append(self.tokenizer.encode(raw_text))
with torch.no_grad():
out_ids = sample_sequence(personality, self.history,
self.tokenizer, self.model, self.args)
self.history.append(out_ids)
self.history = self.history[-(2 * self.args.max_history + 1):]
out_text = self.tokenizer.decode(out_ids, skip_special_tokens=True)
return out_text
inputs=model.inputs,
outputs=[output6]) # , output5, output4, output3, output2, output])
coco_path = '/media/hachreak/Magrathea/datasets/coco/coco'
model_path = ''
epochs = 100
batch_size = 3
input_shape = (320, 320, 3)
output_shape = (10, 10)
action = 'train'
# action = 'evaluate'
# validation dataset
dataset_val = u.get_dataset(coco_path, 'val')
gen_val = prepare(dataset_val, epochs, batch_size, input_shape, output_shape)
# fuu = next(gen_val)
# import ipdb; ipdb.set_trace()
# train dataset
dataset_train = u.get_dataset(coco_path, 'train')
gen_train = prepare(dataset_train, epochs, batch_size, input_shape,
output_shape)
callback = ModelSaveBestAvgAcc(filepath="model-{epoch:02d}-{avgacc:.2f}.hdf5",
verbose=True,
cond=filter_val('fmeasure'))
losses = []
def main(args: argparse.Namespace):
logger = CompleteLogger(args.log, args.phase)
print(args)
if args.seed is not None:
random.seed(args.seed)
torch.manual_seed(args.seed)
cudnn.deterministic = True
warnings.warn('You have chosen to seed training. '
'This will turn on the CUDNN deterministic setting, '
'which can slow down your training considerably! '
'You may see unexpected behavior when restarting '
'from checkpoints.')
cudnn.benchmark = True
# Data loading code
train_transform = utils.get_train_transform(args.train_resizing,
not args.no_hflip,
args.color_jitter)
val_transform = utils.get_val_transform(args.val_resizing)
print("train_transform: ", train_transform)
print("val_transform: ", val_transform)
train_dataset, val_dataset, num_classes = utils.get_dataset(
args.data, args.root, train_transform, val_transform, args.sample_rate,
args.num_samples_per_classes)
train_loader = DataLoader(train_dataset,
batch_size=args.batch_size,
shuffle=True,
num_workers=args.workers,
drop_last=True)
train_iter = ForeverDataIterator(train_loader)
val_loader = DataLoader(val_dataset,
batch_size=args.batch_size,
shuffle=False,
num_workers=args.workers)
print("training dataset size: {} test dataset size: {}".format(
len(train_dataset), len(val_dataset)))
# create model
print("=> using pre-trained model '{}'".format(args.arch))
backbone = utils.get_model(args.arch, args.pretrained)
backbone_source = utils.get_model(args.arch, args.pretrained)
pool_layer = nn.Identity() if args.no_pool else None
classifier = Classifier(backbone,
num_classes,
pool_layer=pool_layer,
finetune=args.finetune).to(device)
source_classifier = Classifier(backbone_source,
num_classes=backbone_source.fc.out_features,
head=backbone_source.copy_head(),
pool_layer=pool_layer).to(device)
for param in source_classifier.parameters():
param.requires_grad = False
source_classifier.eval()
# define optimizer and lr scheduler
optimizer = SGD(classifier.get_parameters(args.lr),
momentum=args.momentum,
weight_decay=args.wd,
nesterov=True)
lr_scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer,
args.lr_decay_epochs,
gamma=args.lr_gamma)
# resume from the best checkpoint
if args.phase == 'test':
checkpoint = torch.load(logger.get_checkpoint_path('best'),
map_location='cpu')
classifier.load_state_dict(checkpoint)
acc1 = utils.validate(val_loader, classifier, args, device)
print(acc1)
return
# create intermediate layer getter
if args.arch == 'resnet50':
return_layers = [
'backbone.layer1.2.conv3', 'backbone.layer2.3.conv3',
'backbone.layer3.5.conv3', 'backbone.layer4.2.conv3'
]
elif args.arch == 'resnet101':
return_layers = [
'backbone.layer1.2.conv3', 'backbone.layer2.3.conv3',
'backbone.layer3.5.conv3', 'backbone.layer4.2.conv3'
]
else:
raise NotImplementedError(args.arch)
source_getter = IntermediateLayerGetter(source_classifier,
return_layers=return_layers)
target_getter = IntermediateLayerGetter(classifier,
return_layers=return_layers)
# get regularization
if args.regularization_type == 'l2_sp':
backbone_regularization = SPRegularization(source_classifier.backbone,
classifier.backbone)
elif args.regularization_type == 'feature_map':
backbone_regularization = BehavioralRegularization()
elif args.regularization_type == 'attention_feature_map':
attention_file = os.path.join(logger.root, args.attention_file)
if not os.path.exists(attention_file):
attention = calculate_channel_attention(train_dataset,
return_layers, num_classes,
args)
torch.save(attention, attention_file)
else:
print("Loading channel attention from", attention_file)
attention = torch.load(attention_file)
attention = [a.to(device) for a in attention]
backbone_regularization = AttentionBehavioralRegularization(attention)
else:
raise NotImplementedError(args.regularization_type)
head_regularization = L2Regularization(
nn.ModuleList([classifier.head, classifier.bottleneck]))
# start training
best_acc1 = 0.0
for epoch in range(args.epochs):
print(lr_scheduler.get_lr())
# train for one epoch
train(train_iter, classifier, backbone_regularization,
head_regularization, target_getter, source_getter, optimizer,
epoch, args)
lr_scheduler.step()
# evaluate on validation set
acc1 = utils.validate(val_loader, classifier, args, device)
# remember best acc@1 and save checkpoint
torch.save(classifier.state_dict(),
logger.get_checkpoint_path('latest'))
if acc1 > best_acc1:
shutil.copy(logger.get_checkpoint_path('latest'),
logger.get_checkpoint_path('best'))
best_acc1 = max(acc1, best_acc1)
print("best_acc1 = {:3.1f}".format(best_acc1))
logger.close()
def main():
config = utils.parse_args()
if config['cuda'] and torch.cuda.is_available():
device = 'cuda:0'
else:
device = 'cpu'
dataset_args = (config['task'], config['dataset'], config['dataset_path'],
'train', config['num_layers'], config['self_loop'],
config['normalize_adj'], config['transductive'])
dataset = utils.get_dataset(dataset_args)
loader = DataLoader(dataset=dataset,
batch_size=config['batch_size'],
shuffle=True,
collate_fn=dataset.collate_wrapper)
input_dim, output_dim = dataset.get_dims()
model = models.GAT(input_dim, config['hidden_dims'], output_dim,
config['num_heads'], config['dropout'], device)
model.to(device)
if not config['load']:
criterion = utils.get_criterion(config['task'])
optimizer = optim.Adam(model.parameters(),
lr=config['lr'],
weight_decay=config['weight_decay'])
epochs = config['epochs']
stats_per_batch = config['stats_per_batch']
num_batches = int(ceil(len(dataset) / config['batch_size']))
model.train()
print('--------------------------------')
print('Training.')
for epoch in range(epochs):
print('Epoch {} / {}'.format(epoch + 1, epochs))
running_loss = 0.0
num_correct, num_examples = 0, 0
for (idx, batch) in enumerate(loader):
features, node_layers, mappings, rows, labels = batch
features, labels = features.to(device), labels.to(device)
optimizer.zero_grad()
out = model(features, node_layers, mappings, rows)
loss = criterion(out, labels)
loss.backward()
optimizer.step()
with torch.no_grad():
running_loss += loss.item()
predictions = torch.max(out, dim=1)[1]
num_correct += torch.sum(predictions == labels).item()
num_examples += len(labels)
if (idx + 1) % stats_per_batch == 0:
running_loss /= stats_per_batch
accuracy = num_correct / num_examples
print(' Batch {} / {}: loss {}, accuracy {}'.format(
idx + 1, num_batches, running_loss, accuracy))
running_loss = 0.0
num_correct, num_examples = 0, 0
print('Finished training.')
print('--------------------------------')
if config['save']:
print('--------------------------------')
directory = os.path.join(os.path.dirname(os.getcwd()),
'trained_models')
if not os.path.exists(directory):
os.makedirs(directory)
fname = utils.get_fname(config)
path = os.path.join(directory, fname)
print('Saving model at {}'.format(path))
torch.save(model.state_dict(), path)
print('Finished saving model.')
print('--------------------------------')
if config['load']:
directory = os.path.join(os.path.dirname(os.getcwd()),
'trained_models')
fname = utils.get_fname(config)
path = os.path.join(directory, fname)
model.load_state_dict(torch.load(path))
dataset_args = (config['task'], config['dataset'], config['dataset_path'],
'test', config['num_layers'], config['self_loop'],
config['normalize_adj'], config['transductive'])
dataset = utils.get_dataset(dataset_args)
loader = DataLoader(dataset=dataset,
batch_size=config['batch_size'],
shuffle=False,
collate_fn=dataset.collate_wrapper)
criterion = utils.get_criterion(config['task'])
stats_per_batch = config['stats_per_batch']
num_batches = int(ceil(len(dataset) / config['batch_size']))
model.eval()
print('--------------------------------')
print('Testing.')
running_loss, total_loss = 0.0, 0.0
num_correct, num_examples = 0, 0
total_correct, total_examples = 0, 0
for (idx, batch) in enumerate(loader):
features, node_layers, mappings, rows, labels = batch
features, labels = features.to(device), labels.to(device)
out = model(features, node_layers, mappings, rows)
loss = criterion(out, labels)
running_loss += loss.item()
total_loss += loss.item()
predictions = torch.max(out, dim=1)[1]
num_correct += torch.sum(predictions == labels).item()
total_correct += torch.sum(predictions == labels).item()
num_examples += len(labels)
total_examples += len(labels)
if (idx + 1) % stats_per_batch == 0:
running_loss /= stats_per_batch
accuracy = num_correct / num_examples
print(' Batch {} / {}: loss {}, accuracy {}'.format(
idx + 1, num_batches, running_loss, accuracy))
running_loss = 0.0
num_correct, num_examples = 0, 0
total_loss /= num_batches
total_accuracy = total_correct / total_examples
print('Loss {}, accuracy {}'.format(total_loss, total_accuracy))
print('Finished testing.')
print('--------------------------------')
def train_student():
#get args
args = get_args()
seed = set_seed(args.seed, args.use_cuda)
trainset, testset, nr_channels, mlp_input_neurons, classes = get_dataset(
args)
trainloader = torch.utils.data.DataLoader(trainset,
batch_size=args.batch_size,
shuffle=True,
num_workers=args.num_processes)
testloader = torch.utils.data.DataLoader(testset,
batch_size=args.batch_size,
shuffle=False,
num_workers=1)
#get student and teacher models
student_model_class = get_model_class(args.student_model)
teacher_model_class = get_model_class(args.teacher_model)
if "MLP" in args.student_model:
stud_model_simple = student_model_class(mlp_input_neurons, 10,
args.dropout)
stud_model_teacher = student_model_class(mlp_input_neurons, 10,
args.dropout)
teacher_model = teacher_model_class(mlp_input_neurons, 10,
args.dropout)
else:
stud_model_simple = student_model_class(nr_channels, 10, args.dropout)
stud_model_teacher = student_model_class(nr_channels, 10, args.dropout)
teacher_model = teacher_model_class(nr_channels, 10, args.dropout)
print("Train student with teacher help")
loss_epoch2, loss_values2, total_accuracy2 = train_student_teacher(
stud_model_teacher, teacher_model, args, trainloader, testloader, seed)
print("Train simple student")
loss_epoch1, loss_values1, total_accuracy1 = train_student_normal(
stud_model_simple, args, trainloader, testloader, seed)
with open(
"params" + args.dataset + '_' + args.teacher_model + '_' +
str(seed), "rb") as f:
_, epoch_eval_teacher, total_accuracy_teacher = pickle.load(f)
#plot loss and total accuracy
plt.figure(1)
plt.plot(range(0, args.nr_epochs), loss_values1)
plt.plot(range(0, args.nr_epochs), loss_values2)
plt.legend(['student_simple', 'student_teacher'], loc='upper right')
plt.xlabel('Nr Epochs')
plt.ylabel('Loss function value')
plt.title('Loss function comparison between students')
plt.savefig('Loss_function_' + args.dataset + '_students' + str(seed) +
"_" + str(args.id))
plt.figure(2)
plt.plot(loss_epoch1, total_accuracy1)
plt.plot(loss_epoch2, total_accuracy2)
plt.plot(epoch_eval_teacher, total_accuracy_teacher)
plt.legend(['student_simple', 'student_teacher', 'teacher'],
loc='lower right')
plt.xlabel('Nr Epochs')
plt.ylabel('Total accuracy')
plt.title('Accuracy comparison between students')
plt.savefig('Accuracy_' + args.dataset + '_students' + str(seed) + "_" +
str(args.id))
def attack_run(model, adversary, hps):
model.eval()
dataset = get_dataset(data_name=hps.problem, train=False)
# hps.n_batch_test = 1
test_loader = DataLoader(dataset=dataset,
batch_size=hps.n_batch_test,
shuffle=False)
test_clnloss = 0
clncorrect = 0
test_advloss = 0
advcorrect = 0
attack_path = os.path.join(hps.attack_dir, hps.attack)
if not os.path.exists(attack_path):
os.mkdir(attack_path)
for batch_id, (clndata, target) in enumerate(test_loader):
# Note that images are scaled to [-1.0, 1.0]
clndata, target = clndata.to(hps.device), target.to(hps.device)
path = os.path.join(attack_path, 'original_{}.png'.format(batch_id))
save_image(clndata, path, normalize=True)
with torch.no_grad():
output = model(clndata)
print('original logits ', output.detach().cpu().numpy())
test_clnloss += F.cross_entropy(output, target, reduction='sum').item()
pred = output.max(1, keepdim=True)[1]
#print('pred: ', pred)
clncorrect += pred.eq(target.view_as(pred)).sum().item()
advdata = adversary.perturb(clndata, target)
path = os.path.join(attack_path,
'{}perturbed_{}.png'.format(prefix, batch_id))
save_image(advdata, path, normalize=True)
with torch.no_grad():
output = model(advdata)
print('adv logits ', output.detach().cpu().numpy())
test_advloss += F.cross_entropy(output, target, reduction='sum').item()
pred = output.max(1, keepdim=True)[1]
#print('pred: ', pred)
advcorrect += pred.eq(target.view_as(pred)).sum().item()
#if batch_id == 2:
# exit(0)
break
test_clnloss /= len(test_loader.dataset)
print('Test set: avg cln loss: {:.4f},'
' cln acc: {}/{}'.format(test_clnloss, clncorrect,
len(test_loader.dataset)))
test_advloss /= len(test_loader.dataset)
print('Test set: avg adv loss: {:.4f},'
' adv acc: {}/{}'.format(test_advloss, advcorrect,
len(test_loader.dataset)))
cln_acc = clncorrect / len(test_loader.dataset)
adv_acc = advcorrect / len(test_loader.dataset)
return cln_acc, adv_acc
from glob import glob
import tensorflow as tf
from tensorflow.keras.callbacks import TensorBoard, ModelCheckpoint
import os
from tfrecord_parser import parse_tfrecords
from tfrecord_creator import create_tfrecords
from utils import get_dataset
print('TensorFlow', tf.__version__)
images_path, mask_path, num_classes, dataset_size = get_dataset(
dataset_path='../mini', folder='training')
batch_size = 1
H, W = 512, 512
tfrecord_dir = os.path.join(os.getcwd(), 'tfrecords')
os.makedirs(tfrecord_dir, exist_ok=True)
checkpoint_dir = os.path.join(os.getcwd(), 'checkpoints')
os.makedirs(checkpoint_dir, exist_ok=True)
# create_tfrecords(images_path, mask_path, tfrecord_dir)
train_tfrecords = os.path.join(tfrecord_dir, 'train*.tfrecord')
input_function = parse_tfrecords(filenames=train_tfrecords,
height=H,
width=W,
batch_size=batch_size)
def attack_run_rejection_policy(model, adversary, hps):
"""
An attack run with rejection policy.
:param model: Pytorch model.
:param adversary: Advertorch adversary.
:param hps: hyperparameters
:return:
"""
model.eval()
# Get thresholds
threshold_list1 = []
threshold_list2 = []
for label_id in range(hps.n_classes):
# No data augmentation(crop_flip=False) when getting in-distribution thresholds
dataset = get_dataset(data_name=hps.problem,
train=True,
label_id=label_id,
crop_flip=False)
in_test_loader = DataLoader(dataset=dataset,
batch_size=hps.n_batch_test,
shuffle=False)
print('Inference on {}, label_id {}'.format(hps.problem, label_id))
in_ll_list = []
for batch_id, (x, y) in enumerate(in_test_loader):
x = x.to(hps.device)
y = y.to(hps.device)
ll = model(x)
correct_idx = ll.argmax(dim=1) == y
ll_, y_ = ll[correct_idx], y[
correct_idx] # choose samples are classified correctly
in_ll_list += list(ll_[:, label_id].detach().cpu().numpy())
thresh_idx = int(0.01 * len(in_ll_list))
thresh1 = sorted(in_ll_list)[thresh_idx]
thresh_idx = int(0.02 * len(in_ll_list))
thresh2 = sorted(in_ll_list)[thresh_idx]
threshold_list1.append(thresh1) # class mean as threshold
threshold_list2.append(thresh2) # class mean as threshold
print('1st & 2nd percentile thresholds: {:.3f}, {:.3f}'.format(
thresh1, thresh2))
# Evaluation
dataset = get_dataset(data_name=hps.problem, train=False)
# hps.n_batch_test = 1
test_loader = DataLoader(dataset=dataset,
batch_size=hps.n_batch_test,
shuffle=False)
n_correct = 0 # total number of correct classified samples by clean classifier
n_successful_adv = 0 # total number of successful adversarial examples generated
n_rejected_adv1 = 0 # total number of successfully rejected (successful) adversarial examples, <= n_successful_adv
n_rejected_adv2 = 0 # total number of successfully rejected (successful) adversarial examples, <= n_successful_adv
attack_path = os.path.join(hps.attack_dir, hps.attack)
if not os.path.exists(attack_path):
os.mkdir(attack_path)
thresholds1 = torch.tensor(threshold_list1).to(hps.device)
thresholds2 = torch.tensor(threshold_list2).to(hps.device)
l2_distortion_list = []
for batch_id, (x, y) in enumerate(test_loader):
# Note that images are scaled to [0., 1.0]
x, y = x.to(hps.device), y.to(hps.device)
with torch.no_grad():
output = model(x)
pred = output.argmax(dim=1)
correct_idx = pred == y
x, y = x[correct_idx], y[
correct_idx] # Only evaluate on the correct classified samples by clean classifier.
n_correct += correct_idx.sum().item()
adv_x = adversary.perturb(x, y)
with torch.no_grad():
output = model(adv_x)
diff = adv_x - x
l2_distortion = diff.norm(p=2,
dim=-1).mean().item() # mean l2 distortion
pred = output.argmax(dim=1)
successful_idx = pred != y # idx of successful adversarial examples.
values, pred = output[successful_idx].max(dim=1)
# confidence_idx = values >= thresholds[pred]
reject_idx1 = values < thresholds1[
pred] # idx of successfully rejected samples.
reject_idx2 = values < thresholds2[
pred] # idx of successfully rejected samples.
# adv_correct += pred[confidence_idx].eq(y[confidence_idx]).sum().item()
n_successful_adv += successful_idx.float().sum().item()
n_rejected_adv1 += reject_idx1.float().sum().item()
n_rejected_adv2 += reject_idx2.float().sum().item()
l2_distortion_list.append(l2_distortion)
if batch_id % 10 == 0:
print('Evaluating on {}-th batch ...'.format(batch_id + 1))
n = len(test_loader.dataset)
reject_rate1 = n_rejected_adv1 / n_successful_adv
reject_rate2 = n_rejected_adv2 / n_successful_adv
success_adv_rate = n_successful_adv / n_correct
print('Test set, clean classification accuracy: {}/{}={:.4f}'.format(
n_correct, n, n_correct / n))
print('success rate of adv examples generation: {}/{}={:.4f}'.format(
n_successful_adv, n_correct, success_adv_rate))
print('Mean L2 distortion of Adv Examples: {:.4f}'.format(
np.mean(l2_distortion_list)))
print('1st percentile, reject success rate: {}/{}={:.4f}'.format(
n_rejected_adv1, n_successful_adv, reject_rate1))
print('2nd percentile, reject success rate: {}/{}={:.4f}'.format(
n_rejected_adv2, n_successful_adv, reject_rate2))
def main(args: argparse.Namespace):
logger = CompleteLogger(args.log, args.phase)
print(args)
if args.seed is not None:
random.seed(args.seed)
torch.manual_seed(args.seed)
cudnn.deterministic = True
warnings.warn('You have chosen to seed training. '
'This will turn on the CUDNN deterministic setting, '
'which can slow down your training considerably! '
'You may see unexpected behavior when restarting '
'from checkpoints.')
cudnn.benchmark = True
# Data loading code
train_transform = utils.get_train_transform(args.train_resizing, random_horizontal_flip=True,
random_color_jitter=False)
val_transform = utils.get_val_transform(args.val_resizing)
print("train_transform: ", train_transform)
print("val_transform: ", val_transform)
train_source_dataset, train_target_dataset, val_dataset, test_dataset, num_classes, args.class_names = \
utils.get_dataset(args.data, args.root, args.source, args.target, train_transform, val_transform)
train_source_loader = DataLoader(train_source_dataset, batch_size=args.batch_size,
shuffle=True, num_workers=args.workers, drop_last=True)
train_target_loader = DataLoader(train_target_dataset, batch_size=args.batch_size,
shuffle=True, num_workers=args.workers, drop_last=True)
val_loader = DataLoader(val_dataset, batch_size=args.batch_size, shuffle=False, num_workers=args.workers)
test_loader = DataLoader(test_dataset, batch_size=args.batch_size, shuffle=False, num_workers=args.workers)
train_source_iter = ForeverDataIterator(train_source_loader)
train_target_iter = ForeverDataIterator(train_target_loader)
# create model
print("=> using pre-trained model '{}'".format(args.arch))
backbone = utils.get_model(args.arch)
pool_layer = nn.Identity() if args.no_pool else None
classifier = ImageClassifier(backbone, num_classes, bottleneck_dim=args.bottleneck_dim, pool_layer=pool_layer).to(device)
domain_discri = DomainDiscriminator(in_feature=classifier.features_dim, hidden_size=1024).to(device)
# define optimizer and lr scheduler
optimizer = SGD(classifier.get_parameters() + domain_discri.get_parameters(),
args.lr, momentum=args.momentum, weight_decay=args.weight_decay, nesterov=True)
lr_scheduler = LambdaLR(optimizer, lambda x: args.lr * (1. + args.lr_gamma * float(x)) ** (-args.lr_decay))
# define loss function
domain_adv = DomainAdversarialLoss(domain_discri).to(device)
# analysis the model
if args.phase == 'analysis':
# extract features from both domains
feature_extractor = nn.Sequential(classifier.backbone, classifier.pool_layer, classifier.bottleneck).to(device)
source_feature = collect_feature(train_source_loader, feature_extractor, device)
target_feature = collect_feature(train_target_loader, feature_extractor, device)
# plot t-SNE
tSNE_filename = osp.join(logger.visualize_directory, 'TSNE.png')
tsne.visualize(source_feature, target_feature, tSNE_filename)
print("Saving t-SNE to", tSNE_filename)
# calculate A-distance, which is a measure for distribution discrepancy
A_distance = a_distance.calculate(source_feature, target_feature, device)
print("A-distance =", A_distance)
return
if args.phase == 'test':
acc1 = validate(test_loader, classifier, args)
print(acc1)
return
# start training
best_h_score = 0.
for epoch in range(args.epochs):
# train for one epoch
train(train_source_iter, train_target_iter, classifier, domain_adv, optimizer,
lr_scheduler, epoch, args)
# evaluate on validation set
h_score = validate(val_loader, classifier, args)
# remember best acc@1 and save checkpoint
torch.save(classifier.state_dict(), logger.get_checkpoint_path('latest'))
if h_score > best_h_score:
shutil.copy(logger.get_checkpoint_path('latest'), logger.get_checkpoint_path('best'))
best_h_score = max(h_score, best_h_score)
print("best_h_score = {:3.1f}".format(best_h_score))
# evaluate on test set
classifier.load_state_dict(torch.load(logger.get_checkpoint_path('best')))
h_score = validate(test_loader, classifier, args)
print("test_h_score = {:3.1f}".format(h_score))
logger.close()
def main(args: argparse.Namespace):
logger = CompleteLogger(args.log, args.phase)
print(args)
if args.seed is not None:
random.seed(args.seed)
torch.manual_seed(args.seed)
cudnn.deterministic = True
warnings.warn('You have chosen to seed training. '
'This will turn on the CUDNN deterministic setting, '
'which can slow down your training considerably! '
'You may see unexpected behavior when restarting '
'from checkpoints.')
cudnn.benchmark = True
# Data loading code
train_transform = utils.get_train_transform(args.train_resizing, random_horizontal_flip=not args.no_hflip,
random_color_jitter=False, resize_size=args.resize_size,
norm_mean=args.norm_mean, norm_std=args.norm_std)
val_transform = utils.get_val_transform(args.val_resizing, resize_size=args.resize_size,
norm_mean=args.norm_mean, norm_std=args.norm_std)
print("train_transform: ", train_transform)
print("val_transform: ", val_transform)
train_source_dataset, train_target_dataset, val_dataset, test_dataset, num_classes, args.class_names = \
utils.get_dataset(args.data, args.root, args.source, args.target, train_transform, val_transform)
train_source_loader = DataLoader(train_source_dataset, batch_size=args.batch_size,
shuffle=True, num_workers=args.workers, drop_last=True)
train_target_loader = DataLoader(train_target_dataset, batch_size=args.batch_size,
shuffle=True, num_workers=args.workers, drop_last=True)
val_loader = DataLoader(val_dataset, batch_size=args.batch_size, shuffle=False, num_workers=args.workers)
test_loader = DataLoader(test_dataset, batch_size=args.batch_size, shuffle=False, num_workers=args.workers)
train_source_iter = ForeverDataIterator(train_source_loader)
train_target_iter = ForeverDataIterator(train_target_loader)
# create model
print("=> using model '{}'".format(args.arch))
backbone = utils.get_model(args.arch, pretrain=not args.scratch)
pool_layer = nn.Identity() if args.no_pool else None
classifier = ImageClassifier(backbone, num_classes, args.num_blocks,
bottleneck_dim=args.bottleneck_dim, dropout_p=args.dropout_p,
pool_layer=pool_layer, finetune=not args.scratch).to(device)
adaptive_feature_norm = AdaptiveFeatureNorm(args.delta).to(device)
# define optimizer
# the learning rate is fixed according to origin paper
optimizer = SGD(classifier.get_parameters(), args.lr, weight_decay=args.weight_decay)
# resume from the best checkpoint
if args.phase != 'train':
checkpoint = torch.load(logger.get_checkpoint_path('best'), map_location='cpu')
classifier.load_state_dict(checkpoint)
# analysis the model
if args.phase == 'analysis':
# extract features from both domains
feature_extractor = nn.Sequential(classifier.backbone, classifier.pool_layer, classifier.bottleneck).to(device)
source_feature = collect_feature(train_source_loader, feature_extractor, device)
target_feature = collect_feature(train_target_loader, feature_extractor, device)
# plot t-SNE
tSNE_filename = osp.join(logger.visualize_directory, 'TSNE.pdf')
tsne.visualize(source_feature, target_feature, tSNE_filename)
print("Saving t-SNE to", tSNE_filename)
# calculate A-distance, which is a measure for distribution discrepancy
A_distance = a_distance.calculate(source_feature, target_feature, device)
print("A-distance =", A_distance)
return
if args.phase == 'test':
acc1 = utils.validate(test_loader, classifier, args, device)
print(acc1)
return
# start training
best_acc1 = 0.
for epoch in range(args.epochs):
# train for one epoch
train(train_source_iter, train_target_iter, classifier, adaptive_feature_norm, optimizer, epoch, args)
# evaluate on validation set
acc1 = utils.validate(val_loader, classifier, args, device)
# remember best acc@1 and save checkpoint
torch.save(classifier.state_dict(), logger.get_checkpoint_path('latest'))
if acc1 > best_acc1:
shutil.copy(logger.get_checkpoint_path('latest'), logger.get_checkpoint_path('best'))
best_acc1 = max(acc1, best_acc1)
print("best_acc1 = {:3.1f}".format(best_acc1))
# evaluate on test set
classifier.load_state_dict(torch.load(logger.get_checkpoint_path('best')))
acc1 = utils.validate(test_loader, classifier, args, device)
print("test_acc1 = {:3.1f}".format(acc1))
logger.close()
'indicator_{}'.format(clean_method))
utils.save_dfs(ind_train, ind_test, ind_path_pfx)
def clean(dataset, error_type=None):
""" Clean each error in the dataset.
Args:
dataset (dict): dataset dict in dataset.py
"""
print("- Clean dataset '{}'".format(dataset['data_dir']))
for error in dataset['error_types']:
if error_type is not None and error != error_type:
continue
print(" - Clean error type '{}'".format(error))
clean_error(dataset, error)
print(" - Finished")
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--dataset', default=None)
args = parser.parse_args()
# datasets to be cleaned, clean all datasets if not specified
datasets = [utils.get_dataset(args.dataset)
] if args.dataset is not None else config.datasets
# clean datasets
for dataset in datasets:
clean(dataset)
import numpy as np
import re
import time
import utils
from bpe import BPE
bpe = BPE()
bpe.load('data/words.bpe')
endToken = bpe.str_to_token['\n']
numTokens = len(bpe.str_to_token)
raw_data = utils.get_dataset(low_it=True)
SAVE_NAME = 'data/train_data_bpe_edition.npz'
DATA_TO_PROCESS = int(input('data to process, {} max? '.format(len(raw_data))))
BATCH_SIZE = int(input('batch size? '))
if DATA_TO_PROCESS > len(raw_data):
DATA_TO_PROCESS = len(raw_data)
vocab = utils.vocab
xi, xp, y = utils.encodeData(raw_data[:DATA_TO_PROCESS], bpe, endToken)
print('to start processing {} of data, press any key'.format(len(xi)))
input()
print('{:=^40}'.format(' starting '))
start = time.time()
def main():
parser = argparse.ArgumentParser(description='TrainingContainer')
parser.add_argument('--algorithm-settings',
type=str,
default="",
help="algorithm settings")
parser.add_argument('--search-space',
type=str,
default="",
help="search space for the neural architecture search")
parser.add_argument('--num-layers',
type=str,
default="",
help="number of layers of the neural network")
args = parser.parse_args()
# Get Algorithm Settings
algorithm_settings = args.algorithm_settings.replace("\'", "\"")
algorithm_settings = json.loads(algorithm_settings)
print(">>> Algorithm settings")
for key, value in algorithm_settings.items():
if len(key) > 13:
print("{}\t{}".format(key, value))
elif len(key) < 5:
print("{}\t\t\t{}".format(key, value))
else:
print("{}\t\t{}".format(key, value))
print()
num_epochs = int(algorithm_settings["num_epochs"])
w_lr = float(algorithm_settings["w_lr"])
w_lr_min = float(algorithm_settings["w_lr_min"])
w_momentum = float(algorithm_settings["w_momentum"])
w_weight_decay = float(algorithm_settings["w_weight_decay"])
w_grad_clip = float(algorithm_settings["w_grad_clip"])
alpha_lr = float(algorithm_settings["alpha_lr"])
alpha_weight_decay = float(algorithm_settings["alpha_weight_decay"])
batch_size = int(algorithm_settings["batch_size"])
num_workers = int(algorithm_settings["num_workers"])
init_channels = int(algorithm_settings["init_channels"])
print_step = int(algorithm_settings["print_step"])
num_nodes = int(algorithm_settings["num_nodes"])
stem_multiplier = int(algorithm_settings["stem_multiplier"])
# Get Search Space
search_space = args.search_space.replace("\'", "\"")
search_space = json.loads(search_space)
search_space = SearchSpace(search_space)
# Get Num Layers
num_layers = int(args.num_layers)
print("Number of layers {}\n".format(num_layers))
# Set GPU Device
# Currently use only first available GPU
# TODO: Add multi GPU support
# TODO: Add functionality to select GPU
all_gpus = list(range(torch.cuda.device_count()))
if len(all_gpus) > 0:
device = torch.device("cuda")
torch.cuda.set_device(all_gpus[0])
np.random.seed(2)
torch.manual_seed(2)
torch.cuda.manual_seed_all(2)
torch.backends.cudnn.benchmark = True
print(">>> Use GPU for Training <<<")
print("Device ID: {}".format(torch.cuda.current_device()))
print("Device name: {}".format(torch.cuda.get_device_name(0)))
print("Device availability: {}\n".format(torch.cuda.is_available()))
else:
device = torch.device("cpu")
print(">>> Use CPU for Training <<<")
# Get dataset with meta information
# TODO: Add support for more dataset
input_channels, num_classes, train_data = utils.get_dataset()
criterion = nn.CrossEntropyLoss().to(device)
model = NetworkCNN(init_channels, input_channels, num_classes, num_layers,
criterion, search_space, num_nodes, stem_multiplier)
model = model.to(device)
# Weights optimizer
w_optim = torch.optim.SGD(model.getWeights(),
w_lr,
momentum=w_momentum,
weight_decay=w_weight_decay)
# Alphas optimizer
alpha_optim = torch.optim.Adam(model.getAlphas(),
alpha_lr,
betas=(0.5, 0.999),
weight_decay=alpha_weight_decay)
# Split data to train/validation
num_train = len(train_data)
split = num_train // 2
indices = list(range(num_train))
train_sampler = torch.utils.data.sampler.SubsetRandomSampler(
indices[:split])
valid_sampler = torch.utils.data.sampler.SubsetRandomSampler(
indices[split:])
train_loader = torch.utils.data.DataLoader(train_data,
batch_size=batch_size,
sampler=train_sampler,
num_workers=num_workers,
pin_memory=True)
valid_loader = torch.utils.data.DataLoader(train_data,
batch_size=batch_size,
sampler=valid_sampler,
num_workers=num_workers,
pin_memory=True)
lr_scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(w_optim,
num_epochs,
eta_min=w_lr_min)
architect = Architect(model, w_momentum, w_weight_decay)
# Start training
best_top1 = 0.
for epoch in range(num_epochs):
lr_scheduler.step()
lr = lr_scheduler.get_lr()[0]
model.print_alphas()
# Training
print(">>> Training")
train(train_loader, valid_loader, model, architect, w_optim,
alpha_optim, lr, epoch, num_epochs, device, w_grad_clip,
print_step)
# Validation
print("\n>>> Validation")
cur_step = (epoch + 1) * len(train_loader)
top1 = validate(valid_loader, model, epoch, cur_step, num_epochs,
device, print_step)
# Print genotype
genotype = model.genotype(search_space)
print("\nModel genotype = {}".format(genotype))
# Modify best top1
if top1 > best_top1:
best_top1 = top1
best_genotype = genotype
print("Final best Prec@1 = {:.4%}".format(best_top1))
print("\nBest-Genotype={}".format(str(best_genotype).replace(" ", "")))
feature_dim, temperature, tau_plus, k = args.feature_dim, args.temperature, args.tau_plus, args.k
batch_size, epochs, estimator = args.batch_size, args.epochs, args.estimator
dataset_name = args.dataset_name
beta = args.beta
anneal = args.anneal
#configuring an adaptive beta if using annealing method
if anneal == 'down':
do_beta_anneal = True
n_steps = 9
betas = iter(np.linspace(beta, 0, n_steps))
else:
do_beta_anneal = False
# data prepare
train_data, memory_data, test_data = utils.get_dataset(dataset_name)
train_loader = DataLoader(train_data,
batch_size=batch_size,
shuffle=True,
num_workers=12,
pin_memory=True,
drop_last=True)
memory_loader = DataLoader(memory_data,
batch_size=batch_size,
shuffle=False,
num_workers=12,
pin_memory=True)
test_loader = DataLoader(test_data,
batch_size=batch_size,
shuffle=False,
def run_training(args):
logging.set_verbosity(logging.WARNING)
args = utils.dict_to_namedtuple(args)
config = hparams_config.get_efficientdet_config(args.model_name)
config.override(args.hparams, allow_new_keys=True)
config.image_size = utils.parse_image_size(config.image_size)
params = dict(
config.as_dict(),
seed=args.seed,
batch_size=args.batch_size,
)
logging.info(params)
if args.ckpt_dir:
ckpt_dir = args.ckpt_dir
if not tf.io.gfile.exists(ckpt_dir):
tf.io.gfile.makedirs(ckpt_dir)
config_file = os.path.join(ckpt_dir, "config.yaml")
if not tf.io.gfile.exists(config_file):
tf.io.gfile.GFile(config_file, "w").write(str(config))
if params["seed"]:
seed = params["seed"]
os.environ["PYTHONHASHSEED"] = str(seed)
tf.random.set_seed(seed)
np.random.seed(seed)
random.seed(seed)
os.environ["TF_DETERMINISTIC_OPS"] = "1"
os.environ["TF_CUDNN_DETERMINISTIC"] = "1"
utils.setup_gpus()
num_devices = 1
physical_devices = tf.config.list_physical_devices("GPU")
multi_gpu = args.multi_gpu
if multi_gpu is not None and len(multi_gpu) != 1 and len(
physical_devices) > 1:
devices = [f"GPU:{gpu}"
for gpu in multi_gpu] if len(multi_gpu) != 0 else None
strategy = tf.distribute.MirroredStrategy(devices)
num_devices = len(devices) if devices else len(physical_devices)
else:
strategy = tf.distribute.get_strategy()
train_dataset = utils.get_dataset(
args,
args.batch_size * num_devices,
True,
params,
strategy if num_devices > 1 else None,
)
if args.eval_after_training or args.eval_during_training:
eval_dataset = utils.get_dataset(
args,
num_devices,
False,
params,
strategy if num_devices > 1 else None,
)
options = tf.data.Options()
options.experimental_distribute.auto_shard_policy = (
tf.data.experimental.AutoShardPolicy.DATA)
eval_dataset = eval_dataset.with_options(options)
with strategy.scope():
model = efficientdet_net.EfficientDetNet(params=params)
global_batch_size = args.batch_size * strategy.num_replicas_in_sync
model.compile(optimizer=optimizers.get_optimizer(
params, args.epochs, global_batch_size, args.train_steps))
initial_epoch = args.initial_epoch
if args.start_weights:
image_size = params["image_size"]
model.predict(np.zeros((1, image_size[0], image_size[1], 3)))
model.load_weights(args.start_weights)
fname = args.start_weights.split("/")[-1]
ckpt_pattern = f"{args.model_name}\.(\d\d+)\.h5"
match = re.match(ckpt_pattern, fname)
if match:
initial_epoch = int(match.group(1).lstrip("0"))
callbacks = []
if args.ckpt_dir:
ckpt_dir = args.ckpt_dir
if not tf.io.gfile.exists(ckpt_dir):
tf.io.gfile.makedirs(tensorboard_dir)
callbacks.append(
tf.keras.callbacks.ModelCheckpoint(
filepath=os.path.join(
ckpt_dir, "".join([args.model_name,
".{epoch:02d}.h5"])),
save_weights_only=True,
))
if args.log_dir:
log_dir = args.log_dir
if not tf.io.gfile.exists(log_dir):
tf.io.gfile.makedirs(log_dir)
callbacks.append(
tf.keras.callbacks.TensorBoard(log_dir=log_dir,
update_freq="epoch"))
model.fit(
train_dataset,
epochs=args.epochs,
steps_per_epoch=args.train_steps,
initial_epoch=initial_epoch,
callbacks=callbacks,
validation_data=eval_dataset
if args.eval_during_training else None,
validation_steps=args.eval_steps,
validation_freq=args.eval_freq,
)
if args.eval_after_training:
print("Evaluation after training:")
model.evaluate(eval_dataset, steps=args.eval_steps)
model.save_weights(args.output_filename)
def main(args: argparse.Namespace):
logger = CompleteLogger(args.log, args.phase)
print(args)
if args.seed is not None:
random.seed(args.seed)
torch.manual_seed(args.seed)
cudnn.deterministic = True
warnings.warn('You have chosen to seed training. '
'This will turn on the CUDNN deterministic setting, '
'which can slow down your training considerably! '
'You may see unexpected behavior when restarting '
'from checkpoints.')
cudnn.benchmark = True
# Data loading code
train_transform = utils.get_train_transform(args.train_resizing, random_horizontal_flip=not args.no_hflip,
random_color_jitter=False, resize_size=args.resize_size,
norm_mean=args.norm_mean, norm_std=args.norm_std)
val_transform = utils.get_val_transform(args.val_resizing, resize_size=args.resize_size,
norm_mean=args.norm_mean, norm_std=args.norm_std)
print("train_transform: ", train_transform)
print("val_transform: ", val_transform)
train_source_dataset, train_target_dataset, val_dataset, test_dataset, num_classes, args.class_names = \
utils.get_dataset(args.data, args.root, args.source, args.target, train_transform, val_transform)
train_source_loader = DataLoader(train_source_dataset, batch_size=args.batch_size,
shuffle=True, num_workers=args.workers, drop_last=True)
train_target_loader = DataLoader(train_target_dataset, batch_size=args.batch_size,
shuffle=True, num_workers=args.workers, drop_last=True)
val_loader = DataLoader(val_dataset, batch_size=args.batch_size, shuffle=False, num_workers=args.workers)
test_loader = DataLoader(test_dataset, batch_size=args.batch_size, shuffle=False, num_workers=args.workers)
train_source_iter = ForeverDataIterator(train_source_loader)
train_target_iter = ForeverDataIterator(train_target_loader)
# create model
print("=> using model '{}'".format(args.arch))
backbone = utils.get_model(args.arch, pretrain=not args.scratch)
pool_layer = nn.Identity() if args.no_pool else None
classifier = ImageClassifier(backbone, num_classes, bottleneck_dim=args.bottleneck_dim,
pool_layer=pool_layer, finetune=not args.scratch).to(device)
# define loss function
if args.adversarial:
thetas = [Theta(dim).to(device) for dim in (classifier.features_dim, num_classes)]
else:
thetas = None
jmmd_loss = JointMultipleKernelMaximumMeanDiscrepancy(
kernels=(
[GaussianKernel(alpha=2 ** k) for k in range(-3, 2)],
(GaussianKernel(sigma=0.92, track_running_stats=False),)
),
linear=args.linear, thetas=thetas
).to(device)
parameters = classifier.get_parameters()
if thetas is not None:
parameters += [{"params": theta.parameters(), 'lr': 0.1} for theta in thetas]
# define optimizer
optimizer = SGD(parameters, args.lr, momentum=args.momentum, weight_decay=args.wd, nesterov=True)
lr_scheduler = LambdaLR(optimizer, lambda x: args.lr * (1. + args.lr_gamma * float(x)) ** (-args.lr_decay))
# resume from the best checkpoint
if args.phase != 'train':
checkpoint = torch.load(logger.get_checkpoint_path('best'), map_location='cpu')
classifier.load_state_dict(checkpoint)
# analysis the model
if args.phase == 'analysis':
# extract features from both domains
feature_extractor = nn.Sequential(classifier.backbone, classifier.pool_layer, classifier.bottleneck).to(device)
source_feature = collect_feature(train_source_loader, feature_extractor, device)
target_feature = collect_feature(train_target_loader, feature_extractor, device)
# plot t-SNE
tSNE_filename = osp.join(logger.visualize_directory, 'TSNE.pdf')
tsne.visualize(source_feature, target_feature, tSNE_filename)
print("Saving t-SNE to", tSNE_filename)
# calculate A-distance, which is a measure for distribution discrepancy
A_distance = a_distance.calculate(source_feature, target_feature, device)
print("A-distance =", A_distance)
return
if args.phase == 'test':
acc1 = utils.validate(test_loader, classifier, args, device)
print(acc1)
return
# start training
best_acc1 = 0.
for epoch in range(args.epochs):
# train for one epoch
train(train_source_iter, train_target_iter, classifier, jmmd_loss, optimizer,
lr_scheduler, epoch, args)
# evaluate on validation set
acc1 = utils.validate(val_loader, classifier, args, device)
# remember best acc@1 and save checkpoint
torch.save(classifier.state_dict(), logger.get_checkpoint_path('latest'))
if acc1 > best_acc1:
shutil.copy(logger.get_checkpoint_path('latest'), logger.get_checkpoint_path('best'))
best_acc1 = max(acc1, best_acc1)
print("best_acc1 = {:3.1f}".format(best_acc1))
# evaluate on test set
classifier.load_state_dict(torch.load(logger.get_checkpoint_path('best')))
acc1 = utils.validate(test_loader, classifier, args, device)
print("test_acc1 = {:3.1f}".format(acc1))
logger.close()
if __name__ == '__main__':
start_time = time.time()
# define paths
path_project = os.path.abspath('..')
logger = SummaryWriter('../logs')
args = args_parser()
exp_details(args)
if args.gpu:
torch.cuda.set_device(args.gpu)
device = 'cuda' if args.gpu else 'cpu'
# load dataset and user groups
train_dataset, test_dataset, user_groups = get_dataset(args)
# BUILD MODEL
if args.model == 'cnn':
# Convolutional neural netork
if args.dataset == 'mnist':
global_model = CNNMnist(args=args)
elif args.dataset == 'fmnist':
global_model = CNNFashion_Mnist(args=args)
elif args.dataset == 'cifar':
global_model = CNNCifar(args=args)
elif args.model == 'mlp':
# Multi-layer preceptron
img_size = train_dataset[0][0].shape
len_in = 1
def user_annotation():
"""
Feed tweets to console one at a time, and ask user for sentiment annotation.
"""
dataset = utils.select_dataset()
text_tweets = utils.get_dataset(dataset)
tweets = []
for text_tweet in text_tweets:
tweets.append(tweet.to_tweet(text_tweet))
username = raw_input("Name? ... ")
print "\n--------------\n"
print "Input: "
print "\n1: Negative sentiment (Negative opinion). \n2: Neutral/objective sentiment (No opinion). \n3: Positive sentiment (Positive opinion). \n5: Delete the tweet from the dataset. \nx: Cancel sequence. 0: Go back to previous tweet. "
print "\n--------------\n"
annotated_to = 0
i = 0
while i < len(tweets):
# tweets[i].text.encode('utf8')
# text = tweets[i].text
# tweets[i].text = text.decode('utf8')
try:
print "Tweet nr. : "+str(i+1)
print str(((i+1.0*1.0)/len(tweets)*1.0)*100)+" % done "
print unicode(tweets[i].__str__().decode('utf8'))
except UnicodeEncodeError:
try:
print "Tweet nr. : "+str(i+1)
print str(tweets[i])
except UnicodeEncodeError:
print "Could not print tweet number "+str(i+1) +". Deleting tweet..."
tweets.remove(tweets[i])
continue
userinput = raw_input("...")
while not legal_input(userinput):
userinput = raw_input("Unlawful input! Please re-introduce.")
if userinput is '1':
tweets[i].set_sentiment("negative")
elif userinput is '2':
tweets[i].set_sentiment("neutral")
elif userinput is '3':
tweets[i].set_sentiment("positive")
elif userinput is '5':
print "Deleting tweet..."
tweets.remove(tweets[i])
continue
elif userinput is '0':
i = i-1
continue
elif userinput is 'x':
break
i = i+1
#TODO: need to encode to utf when getting from dataset?!?!
#Store the sentiment in file!
tweetlines = []
for t in tweets[:i]:
if t.get_sentiment() is None:
continue
tweetlines.append(t.to_tsv())
dir = username+"_annotated_data"
if not os.path.exists(dir):
os.makedirs(dir)
utils.store_dataset(tweetlines, dir+dataset[4:])
print "Domo arigato!"
if __name__ == '__main__':
start_time = time.time()
# define paths
path_project = os.path.abspath('..')
logger = SummaryWriter('../brats2018_logs')
args = brats2018_args_parser()
exp_details(args)
device = torch.device(args.gpu) if args.gpu is not None else 'cpu'
if args.gpu:
torch.cuda.set_device(device)
# load dataset and user groups
train_dataset, _, user_groups = get_dataset(args)
# 使用 UNET
if args.model == 'unet':
global_model = UNet(n_channels=1, n_classes=1, bilinear=True)
else:
exit('Error: unrecognized model')
# Set the model to train and send it to device.
global_model.to(device)
global_model.train()
print(global_model)
# copy weights
global_weights = global_model.state_dict()
import matplotlib.pyplot as plt
import torch
from torch.utils.data import DataLoader
from utils import get_dataset
from options import args_parser
from update import test_inference
from models import MLP, CNNMnist, CNNFashion_Mnist, CNNCifar
if __name__ == '__main__':
args = args_parser()
device = 'cuda' if torch.cuda.is_available() else 'cpu'
# load datasets
train_dataset, test_dataset, _ = get_dataset(args)
# BUILD MODEL
if args.model == 'cnn':
# Convolutional neural netorks
if args.dataset == 'mnist':
global_model = CNNMnist(args=args)
elif args.dataset == 'fmnist':
global_model = CNNFashion_Mnist(args=args)
elif args.dataset == 'cifar':
global_model = CNNCifar(args=args)
elif args.model == 'mlp':
# Multi-layer preceptron
img_size = train_dataset[0][0].shape
len_in = 1
for x in img_size:
from sklearn import cross_validation
from sklearn import grid_search
from plot import plot_boundary
from utils import get_dataset
from utils import get_random_state
from utils import compare
if __name__ == "__main__":
# (Question 1) dt.py: Decision tree
SAMPLE_NUMBER = 2000
TRAIN_SET_SAMPLE_NUM = 150
X, y = get_dataset(SAMPLE_NUMBER)
X_train, y_train = X[:TRAIN_SET_SAMPLE_NUM], y[:TRAIN_SET_SAMPLE_NUM]
X_test, y_test = X[TRAIN_SET_SAMPLE_NUM:], y[TRAIN_SET_SAMPLE_NUM:]
# 1.
decisionTreeClassifier = DecisionTreeClassifier(random_state=get_random_state())
decisionTreeClassifier.fit(X_train, y_train)
y_dtc = decisionTreeClassifier.predict(X_test)
# Plot
plot_boundary("1-1-Ground-Truth", decisionTreeClassifier, X_test, y_test, title="Ground Truth data")
plot_boundary("1-1-Prediction", decisionTreeClassifier, X_test, y_dtc, title="Prediction data")
# 2.
max_depths = [i for i in range(1, 20)]
def main():
parser = argparse.ArgumentParser(description='-----[CNN-classifier]-----')
parser.add_argument(
'--mode',
default='train',
help='train: train (with test) a model / test: test saved models')
parser.add_argument(
'--model',
default='non-static',
help='available models: rand, static, non-static, multichannel')
parser.add_argument('--dataset',
default='TREC',
help='available datasets: MR, TREC, AG, SST1, SST2')
parser.add_argument('--save_model',
default=False,
action='store_true',
help='whether saving model or not')
parser.add_argument('--early_stopping',
default=False,
action='store_true',
help='whether to apply early stopping')
parser.add_argument('--epoch',
default=100,
type=int,
help='number of max epoch')
parser.add_argument('--learning_rate',
default=1.0,
type=float,
help='learning rate')
parser.add_argument('--gpu',
default=-1,
type=int,
help='the number of gpu to be used')
parser.add_argument('--cv',
default=False,
action='store_true',
help='whether to use cross validation')
parser.add_argument(
'--we',
default='w2v',
help='available word embedding: w2v, rand, scv, yelpfscv')
parser.add_argument('--type',
default='CNN',
help='available type for cnn model: CNN, CNN2')
parser.add_argument('--batch_size',
default=50,
type=int,
help='batch_size')
parser.add_argument('--filters',
default=[3, 4, 5],
type=int,
nargs='+',
help='filters')
parser.add_argument('--filter_num',
default=[100, 100, 100],
type=int,
nargs='+',
help='filter_num')
parser.add_argument('--dropout_prob',
default=0.5,
type=float,
help='dropout_prob')
parser.add_argument('--norm_limit',
default=3,
type=float,
help='norm_limit')
parser.add_argument('--result_path',
default='result/result_auto.csv',
help='result_path')
options = parser.parse_args()
data = utils.get_dataset(options.dataset)
# data = utils.get_dataset('MR')
params = {
'MODEL':
options.model,
'DATASET':
options.dataset,
'SAVE_MODEL':
options.save_model,
'EARLY_STOPPING':
options.early_stopping,
'EPOCH':
options.epoch,
'LEARNING_RATE':
options.learning_rate,
'MAX_SENT_LEN':
max([
len(sent)
for sent in data['train_x'] + data['dev_x'] + data['test_x']
]),
'BATCH_SIZE':
options.batch_size,
'WORD_DIM':
300,
'VOCAB_SIZE':
len(data['vocab']),
'CLASS_SIZE':
len(data['classes']),
'FILTERS':
options.filters,
'FILTER_NUM':
options.filter_num,
'DROPOUT_PROB':
options.dropout_prob,
'NORM_LIMIT':
options.norm_limit,
'GPU':
options.gpu,
'cv':
options.cv,
'we':
options.we,
'type':
options.type,
'result_path':
options.result_path
}
print('=' * 20 + 'INFORMATION' + '=' * 20)
print('MODEL:', params['MODEL'])
print('DATASET:', params['DATASET'])
print('VOCAB_SIZE:', params['VOCAB_SIZE'])
print('EPOCH:', params['EPOCH'])
print('LEARNING_RATE:', params['LEARNING_RATE'])
print('EARLY_STOPPING:', params['EARLY_STOPPING'])
print('SAVE_MODEL:', params['SAVE_MODEL'])
print('WORD EMBEDDING:', params['we'])
print('MODEL TYPE:', params['type'])
print('BATCH_SIZE', params['BATCH_SIZE'])
print('FILTERS', params['FILTERS'])
print('FILTER_NUM', params['FILTER_NUM'])
print('DROPOUT_PROB', params['DROPOUT_PROB'])
print('NORM_LIMIT', params['NORM_LIMIT'])
print('=' * 20 + 'INFORMATION' + '=' * 20)
if options.mode == 'train':
print('=' * 20 + 'TRAINING STARTED' + '=' * 20)
if params['cv']:
kf = KFold(n_splits=10, shuffle=True)
for index, (train_index,
test_index) in enumerate(kf.split(data['x'])):
print('cv {}'.format(index))
train_index = shuffle(train_index)
dev_size = len(train_index) // 10
dev_index = train_index[:dev_size]
train_index = train_index[dev_size:]
data['dev_x'] = list(itemgetter(*dev_index)(data['x']))
data['train_x'] = list(itemgetter(*train_index)(data['x']))
data['test_x'] = list(itemgetter(*test_index)(data['x']))
data['dev_y'] = list(itemgetter(*dev_index)(data['y']))
data['train_y'] = list(itemgetter(*train_index)(data['y']))
data['test_y'] = list(itemgetter(*test_index)(data['y']))
train(data, params)
else:
model = train(data, params)
if params['SAVE_MODEL']:
utils.save_model(model, params)
print('=' * 20 + 'TRAINING FINISHED' + '=' * 20)
else:
if params['GPU'] == -1:
model = utils.load_model(params)
else:
model = utils.load_model(params).cuda(params['GPU'])
test_acc = test(data, model, params)
print('test acc:', test_acc)
def run(args: DictConfig) -> None:
# cuda_available = torch.cuda.is_available()
torch.manual_seed(args.seed)
torch.cuda.manual_seed_all(args.seed)
# device = "cuda" if cuda_available and args.device == 'cuda' else "cpu"
classifier = eval(args.classifier_name)(args.width,
args.n_classes).to(args.device)
logger.info('Classifier: {}, width: {}, # parameters: {}'.format(
args.classifier_name, args.width, cal_parameters(classifier)))
data_dir = hydra.utils.to_absolute_path(args.data_dir)
train_data = get_dataset(data_name=args.dataset,
data_dir=data_dir,
train=True,
crop_flip=True)
test_data = get_dataset(data_name=args.dataset,
data_dir=data_dir,
train=False,
crop_flip=False)
test_loader = DataLoader(dataset=test_data,
batch_size=args.n_batch_test,
shuffle=False)
optimizer = SGD(classifier.parameters(),
lr=args.lr_max,
momentum=args.momentum,
weight_decay=args.weight_decay)
def run_forward(scheduler):
optimal_loss = 1e5
for epoch in range(1, args.n_epochs + 1):
loss, acc = train_epoch(classifier,
train_loader,
args,
optimizer,
scheduler=scheduler)
if loss < optimal_loss:
optimal_loss = loss
torch.save(classifier.state_dict(), checkpoint)
logger.info(
'Epoch {}, lr: {:.4f}, loss: {:.4f}, acc: {:.4f}'.format(
epoch,
scheduler.get_lr()[0], loss, acc))
if args.adv_generation:
checkpoint = '{}_w{}_at_fast.pth'.format(args.classifier_name,
args.width)
train_loader = DataLoader(dataset=train_data,
batch_size=args.n_batch_train,
shuffle=True)
lr_steps = args.n_epochs * len(train_loader)
scheduler = lr_scheduler.CyclicLR(optimizer,
base_lr=args.lr_min,
max_lr=args.lr_max,
step_size_up=lr_steps / 2,
step_size_down=lr_steps / 2)
run_forward(scheduler)
clean_loss, clean_acc = eval_epoch(classifier,
test_loader,
args,
adversarial=False)
adv_loss, adv_acc = eval_epoch(classifier,
test_loader,
args,
adversarial=True,
save=True)
logger.info('Clean loss: {:.4f}, acc: {:.4f}'.format(
clean_loss, clean_acc))
logger.info('Adversarial loss: {:.4f}, acc: {:.4f}'.format(
adv_loss, adv_acc))
else:
n = len(train_data)
split_size = n // args.n_split
lengths = [split_size] * (args.n_split - 1) + [
n % split_size + split_size
]
datasets_list = random_split(train_data, lengths=lengths)
for split_id, dataset in enumerate(datasets_list):
checkpoint = '{}_w{}_split{}_at_fast.pth'.format(
args.classifier_name, args.width, split_id)
logger.info('Running on subset {}, size: {}'.format(
split_id + 1, len(dataset)))
train_loader = DataLoader(dataset=dataset,
batch_size=args.n_batch_train,
shuffle=True)
lr_steps = args.n_epochs * len(train_loader)
scheduler = lr_scheduler.CyclicLR(optimizer,
base_lr=args.lr_min,
max_lr=args.lr_max,
step_size_up=lr_steps / 2,
step_size_down=lr_steps / 2)
run_forward(scheduler)
clean_loss, clean_acc = eval_epoch(classifier,
test_loader,
args,
adversarial=False)
adv_loss, adv_acc = eval_epoch(classifier,
test_loader,
args,
adversarial=True)
logger.info('Clean loss: {:.4f}, acc: {:.4f}'.format(
clean_loss, clean_acc))
logger.info('Adversarial loss: {:.4f}, acc: {:.4f}'.format(
adv_loss, adv_acc))
def main(args):
#network = importlib.import_module(args.model_def)
subdir = datetime.strftime(datetime.now(), '%Y%m%d-%H%M%S')
log_dir = os.path.join(os.path.expanduser(args.logs_base_dir), subdir)
if not os.path.isdir(
log_dir): # Create the log directory if it doesn't exist
os.makedirs(log_dir)
model_dir = os.path.join(os.path.expanduser(args.models_base_dir), subdir)
if not os.path.isdir(
model_dir): # Create the model directory if it doesn't exist
os.makedirs(model_dir)
# Write arguments to a text file
utils.write_arguments_to_file(args, os.path.join(log_dir, 'arguments.txt'))
# Store some git revision info in a text file in the log directory
src_path, _ = os.path.split(os.path.realpath(__file__))
utils.store_revision_info(src_path, log_dir, ' '.join(sys.argv))
np.random.seed(seed=args.seed)
train_set = utils.get_dataset(args.data_dir)
nrof_classes = len(train_set)
print('nrof_classes: ', nrof_classes)
image_list, label_list = utils.get_image_paths_and_labels(train_set)
image_list = np.array(image_list)
print('total images: {}'.format(len(image_list)))
label_list = np.array(label_list, dtype=np.int32)
dataset_size = len(image_list)
data_reader = DataGenerator(image_list, label_list, args.batch_size)
print('Model directory: %s' % model_dir)
print('Log directory: %s' % log_dir)
if args.pretrained_model:
print('Pre-trained model: %s' %
os.path.expanduser(args.pretrained_model))
with tf.Graph().as_default():
tf.set_random_seed(args.seed)
global_step = tf.Variable(0, trainable=False, name='global_step')
# Placeholder for the learning rate
learning_rate_placeholder = tf.placeholder(tf.float32,
name='learning_rate')
images_placeholder = tf.placeholder(tf.float32, [None, 112, 96, 3],
name='images_placeholder')
labels_placeholder = tf.placeholder(tf.int32, [None],
name='labels_placeholder')
phase_train_placeholder = tf.placeholder(tf.bool, name='phase_train')
learning_rate = tf.train.exponential_decay(
learning_rate_placeholder,
global_step,
args.learning_rate_decay_epochs * args.epoch_size,
args.learning_rate_decay_factor,
staircase=True)
tf.summary.scalar('learning_rate', learning_rate)
print('Using optimizer: {}'.format(args.optimizer))
if args.optimizer == 'ADAGRAD':
opt = tf.train.AdagradOptimizer(learning_rate)
elif args.optimizer == 'MOM':
opt = tf.train.MomentumOptimizer(learning_rate, 0.9)
if args.network == 'sphere_network':
prelogits = network.infer(images_placeholder)
else:
raise Exception('Not supported network: {}'.format(args.loss_type))
if args.loss_type == 'softmax':
cross_entropy_mean = utils.softmax_loss(prelogits,
labels_placeholder,
len(train_set),
args.weight_decay, False)
regularization_losses = tf.get_collection(
tf.GraphKeys.REGULARIZATION_LOSSES)
#loss = cross_entropy_mean + args.weight_decay*tf.add_n(regularization_losses)
loss = cross_entropy_mean + args.weight_decay * tf.add_n(
regularization_losses)
#loss = cross_entropy_mean
else:
raise Exception('Not supported loss type: {}'.format(
args.loss_type))
#loss = tf.add_n([cross_entropy_mean] + regularization_losses, name='total_loss')
losses = {}
losses['total_loss'] = loss
losses['softmax_loss'] = cross_entropy_mean
debug_info = {}
debug_info['prelogits'] = prelogits
grads = opt.compute_gradients(loss, tf.trainable_variables())
train_op = opt.apply_gradients(grads, global_step=global_step)
#save_vars = [var for var in tf.global_variables() if 'Adagrad' not in var.name and 'global_step' not in var.name]
save_vars = tf.global_variables()
#saver = tf.train.Saver(tf.trainable_variables(), max_to_keep=3)
saver = tf.train.Saver(save_vars, max_to_keep=3)
# Build the summary operation based on the TF collection of Summaries.
# Start running operations on the Graph.
gpu_options = tf.GPUOptions(
per_process_gpu_memory_fraction=args.gpu_memory_fraction)
sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options,
allow_soft_placement=True))
# Initialize variables
sess.run(tf.global_variables_initializer(),
feed_dict={phase_train_placeholder: True})
sess.run(tf.local_variables_initializer(),
feed_dict={phase_train_placeholder: True})
with sess.as_default():
#pdb.set_trace()
if args.pretrained_model:
print('Restoring pretrained model: %s' % args.pretrained_model)
saver.restore(sess, os.path.expanduser(args.pretrained_model))
# Training and validation loop
epoch = 0
while epoch < args.max_nrof_epochs:
step = sess.run(global_step, feed_dict=None)
epoch = step // args.epoch_size
# Train for one epoch
train(args, sess, epoch, images_placeholder,
labels_placeholder, data_reader, debug,
learning_rate_placeholder, global_step, losses, train_op,
args.learning_rate_schedule_file)
# Save variables and the metagraph if it doesn't exist already
model_dir = args.models_base_dir
checkpoint_path = os.path.join(model_dir,
'model-%s.ckpt' % 'softmax')
saver.save(sess,
checkpoint_path,
global_step=step,
write_meta_graph=False)
# Evaluate on LFW
return model_dir
