def Start(self):
'''{'''
if not self.__InitialiseConf() is True:
return None
'''}'''
''''''
self.__LogStart()
destination = self.configs['destination'] if 'destination' in self.configs else None
DestDirs = self.configs['DestDirs'] if 'DestDirs' in self.configs else None
DirsFiles = self.configs['DirsFiles'] if 'DirsFiles' in self.configs else None
saver = Saver(destination);
if(saver.CreateDestDirs(DestDirs) is None):
self.__LogMsg(saver.msg)
else:
for destdir in DestDirs:
if(saver.CopyFiles(destdir,DirsFiles[destdir]) is None):
self.__LogMsg(saver.msg)
saver.msg = ""
''''''
self.__LogEnd()
return;
def main(argv):
global PYBICO_VERBOSE
try:
opts, args = getopt.getopt(argv, "hvl:s:i:e:u:p:", ["help"])
except getopt.GetoptError as err:
print(str(err))
usage()
sys.exit(2)
PYBICO_VERBOSE = False
load_format = "txt"
save_format = "xlsx"
load_filename = ""
save_filename = ""
password_path = ""
user = ""
for o, a in opts:
if o == "-v":
PYBICO_VERBOSE = True
elif o in ("-h", "--help"):
usage()
sys.exit()
elif o == "-u":
user = a
elif o == "-p":
password_path = a
elif o == "-l":
load_filename = a
elif o == "-s":
save_filename = a
elif o == "-i":
load_format = a
elif o == "-e":
save_format = a
else:
assert False, "unhandled option"
f = open(password_path, 'r')
password = f.read().strip('\n')
db = DB(user, password)
if load_filename != "":
l = Loader()
data = l.load(load_format, load_filename)
db.add(data)
if save_filename != "":
data = db.get()
s = Saver()
s.save(data, save_format, save_filename)
class Scrapper:
def __init__(self, start_url, savefile, max_depth=10, max_width=100):
"""
max_depth: maximum recursion depth to follow for each link
max_width: maximum number of dict keys, i.e. width of tree
"""
self.parser = Parser()
self.start_url = start_url
self.saver = Saver(savefile, max_width)
self.max_depth = max_depth
self.saver.starting_url(self.start_url)
def start_scrapping(self, depth=0, start_url = None):
if depth == self.max_depth:
return
if start_url == None:
start_url = self.start_url
nested_urls = self.get_urls(start_url)
bool_break = self.save_data(start_url, nested_urls)
if not bool_break:
exit()
if nested_urls == None:
return
else:
for url in nested_urls:
self.start_scrapping(depth+1, url)
def get_urls(self, url):
try:
response = requests.get(url)
web_page = response.content
urls = self.parser.get_links(web_page)
except requests.exceptions.RequestException as re:
print(re)
return
return urls
def save_data(self, start_url, nested_urls):
reply = self.saver.save(start_url, nested_urls)
if not reply:
return False
else:
return True
def __init__(self, start_url, savefile, max_depth=10, max_width=100):
"""
max_depth: maximum recursion depth to follow for each link
max_width: maximum number of dict keys, i.e. width of tree
"""
self.parser = Parser()
self.start_url = start_url
self.saver = Saver(savefile, max_width)
self.max_depth = max_depth
self.saver.starting_url(self.start_url)
def read(folder):
log.info('Reading pretrained network from {}'.format(folder))
saver = Saver(folder)
ckpt_info = saver.get_ckpt_info()
model_opt = ckpt_info['model_opt']
ckpt_fname = ckpt_info['ckpt_fname']
model_id = ckpt_info['model_id']
m = model.get_model(model_opt)
cnn_nlayers = len(model_opt['cnn_filter_size'])
mlp_nlayers = 1
timespan = 1
weights = {}
sess = tf.Session()
saver.restore(sess, ckpt_fname)
output_list = []
for net, nlayers in zip(['cnn', 'mlp'], [cnn_nlayers, mlp_nlayers]):
for ii in xrange(nlayers):
for w in ['w', 'b']:
key = '{}_{}_{}'.format(net, w, ii)
log.info(key)
output_list.append(key)
if net == 'cnn':
for tt in xrange(timespan):
for w in ['beta', 'gamma']:
key = '{}_{}_{}_{}'.format(net, ii, tt, w)
log.info(key)
output_list.append(key)
output_var = []
for key in output_list:
output_var.append(m[key])
output_var_value = sess.run(output_var)
for key, value in zip(output_list, output_var_value):
weights[key] = value
log.info(key)
log.info(value.shape)
return weights
def __init__(self, session_params, ax_interactive=None):
# incorporate kwargs
self.params = session_params
self.__dict__.update(self.params)
self.verify_params()
# sync
self.sync_flag = multiprocessing.Value('b', False)
self.sync_to_save = multiprocessing.Queue()
# saver
self.saver = Saver(self.subj, self.name, self, sync_flag=self.sync_flag)
# hardware
self.cam = PSEye(sync_flag=self.sync_flag, **self.cam_params)
self.ar = AnalogReader(saver_obj_buffer=self.saver.buf, sync_flag=self.sync_flag, **self.ar_params)
# communication
self.ni = NI845x(i2c_on=self.imaging)
# interactivity
self.ax_interactive = ax_interactive
# runtime variables
self.notes = {}
self.mask_idx = -1 #for reselecting mask
self.session_on = 0
self.on = False
self.session_complete = False
self.session_kill = False
self.trial_flag = False
self.trial_on = 0
self.trial_off = 0
self.trial_idx = -1
self.stim_cycle_idx = 0
self.paused = False
self.deliver_override = False
self.roi_pts = None
self.eyelid_buffer = np.zeros(self.eyelid_buffer_size)-1
self.eyelid_buffer_ts = np.zeros(self.eyelid_buffer_size)-1
self.past_flag = False
# sync
self.sync_flag.value = True #trigger all processes to get time
self.sync_val = now() #get this process's time
procs = dict(saver=self.saver, cam=self.cam.pseye, ar=self.ar)
sync_vals = {o:procs[o].sync_val.value for o in procs} #collect all process times
sync_vals['session'] = self.sync_val
self.sync_to_save.put(sync_vals)
# more runtime, anything that must occur after sync
_,self.im = self.cam.get()
def main():
args = parse_args()
saver = Saver(args.model_dir)
model = SELUNet()
with warnings.catch_warnings():
warnings.simplefilter('ignore')
if args.use_cuda:
model = model.cuda()
model, _, params_dict = saver.load_checkpoint(
model, file_name=args.model_name)
model.eval()
filespan = args.filespan
idr_params, _, _ = get_optimal_params(model,
args.valspeechfolder,
args.valpeaksfolder,
args.window,
args.stride,
filespan,
numfiles=60,
use_cuda=False,
thlist=[
0.15, 0.2, 0.25, 0.3, 0.35, 0.4,
0.45, 0.5, 0.55, 0.6, 0.65, 0.7,
0.75
],
spblist=[25],
hctlist=[10, 15, 20, 25, 30])
thr = idr_params['thr']
spb = idr_params['spb']
hct = idr_params['hct']
with open('test_idr.txt', 'w') as f:
print(
'Optimal Hyperparameters\nThreshold: {} Samples Per Bin: {} Histogram Count Threshold: {}'
.format(thr, spb, hct),
file=f,
flush=True)
numfiles = len(glob(os.path.join(args.speechfolder, '*.npy')))
print('Models and Files Loaded')
metrics_list = []
for i in range(0, numfiles, filespan):
if (i + filespan) > numfiles:
break
speech_windowed_data, peak_distance, peak_indicator, indices, actual_gci_locations = create_dataset(
args.speechfolder, args.peaksfolder, args.window, args.stride,
slice(i, i + filespan))
input = to_variable(
th.from_numpy(
np.expand_dims(speech_windowed_data, 1).astype(np.float32)),
args.use_cuda, True)
with warnings.catch_warnings():
prediction = model(input)
predicted_peak_indicator = F.sigmoid(prediction[:, 1]).data.numpy()
predicted_peak_distance = (prediction[:,
0]).data.numpy().astype(np.int32)
predicted_peak_indicator_indices = predicted_peak_indicator > args.threshold
predicted_peak_indicator = predicted_peak_indicator[
predicted_peak_indicator_indices].ravel()
predicted_peak_distance = predicted_peak_distance[
predicted_peak_indicator_indices].ravel()
indices = indices[predicted_peak_indicator_indices]
assert (len(indices) == len(predicted_peak_distance))
assert (len(predicted_peak_distance) == len(predicted_peak_indicator))
positive_distance_indices = predicted_peak_distance < args.window
positive_peak_distances = predicted_peak_distance[
positive_distance_indices]
postive_predicted_peak_indicator = predicted_peak_indicator[
positive_distance_indices]
gci_locations = [
indices[i, d] for i, d in enumerate(positive_peak_distances)
]
locations_true = np.nonzero(actual_gci_locations)[0]
xaxes = np.zeros(len(actual_gci_locations))
xaxes[locations_true] = 1
ground_truth = np.row_stack(
(np.arange(len(actual_gci_locations)), xaxes))
predicted_truth = np.row_stack(
(gci_locations, postive_predicted_peak_indicator))
gx = ground_truth[0, :]
gy = ground_truth[1, :]
px = predicted_truth[0, :]
py = predicted_truth[1, :]
fs = 16000
gci = np.array(
cluster(px,
py,
threshold=thr,
samples_per_bin=spb,
histogram_count_threshold=hct))
predicted_gci_time = gci / fs
target_gci_time = np.nonzero(gy)[0] / fs
gci = np.round(gci).astype(np.int64)
gcilocs = np.zeros_like(gx)
gcilocs[gci] = 1
metric = corrected_naylor_metrics(target_gci_time, predicted_gci_time)
print(metric)
metrics_list.append(metric)
idr = np.mean([
v for m in metrics_list for k, v in m.items()
if k == 'identification_rate'
])
mr = np.mean(
[v for m in metrics_list for k, v in m.items() if k == 'miss_rate'])
far = np.mean([
v for m in metrics_list for k, v in m.items()
if k == 'false_alarm_rate'
])
se = np.mean([
v for m in metrics_list for k, v in m.items()
if k == 'identification_accuracy'
])
print('IDR: {:.5f} MR: {:.5f} FAR: {:.5f} IDA: {:.5f}'.format(
idr, mr, far, se))
with open('test_idr.txt', 'a') as f:
f.write('IDR: {:.5f} MR: {:.5f} FAR: {:.5f} IDA: {:.5f}\n'.format(
idr, mr, far, se))
class Logger:
'''
Class for keep a track of what's going on.
TODO: methods docstrings
TODO: fix logs timestamp as int not str
'''
def __init__(self, debug=False):
self.debug = debug
self.saver = Saver()
self.last_tweets = self.maybe_load(FileNames.last_tweets)
self.giphy_keys = self.maybe_load(FileNames.giphy_keys)
def maybe_load(self, fname):
try:
return load_json(fname)
except:
return {}
def save_dict(self, d, fname):
save_json(d, fname)
self.saver.sync()
def save_last_tweets(self):
self.save_dict(self.last_tweets, FileNames.last_tweets)
def save_giphy_keys(self):
self.save_dict(self.giphy_keys, FileNames.giphy_keys)
def update_last_tweet(self, action, status):
if not action in self.last_tweets.keys():
self.last_tweets[action] = {}
self.last_tweets[action][
'datetime'] = datetime.now().strftime('%Y%m%d%H%M%S')
self.last_tweets[action]['id'] = status.id
self.save_last_tweets()
def tweeted_gif(self, key):
if self.giphy_keys == {}:
self.giphy_keys['keys'] = []
return False
else:
if key in self.giphy_keys['keys']:
return True
else:
self.giphy_keys['keys'].append(key)
self.save_giphy_keys()
return False
def last_action_id(self, action):
try:
return self.last_tweets[action]['id']
except Exception:
return None
def last_action_past_seconds(self, action):
t = datetime.utcfromtimestamp(0)
try:
t = datetime.strptime(
self.last_tweets[action]['datetime'], '%Y%m%d%H%M%S')
except Exception, e:
self.log(
'Reading self.last_tweets[action]["datetime"]' + str(e), error=True)
return (datetime.now() - t).total_seconds()
class TrainModel:
def __init__(self):
# prepare training parameters
if args.backbone == "vgg16":
model = torchvision.models.vgg16_bn(True)
# remove dropout
model.classifier = nn.Sequential(
*[model.classifier[i] for i in [0, 1, 3, 4, 6]])
mean = [0.485, 0.456, 0.406]
std = [0.229, 0.224, 0.225]
else:
raise NotImplementedError()
criterion_ce = nn.CrossEntropyLoss()
if args.loss_type == "PC1_sign":
criterion_corr = PearsonCorrelationLoss1("sign")
elif args.loss_type == "PC2_sign":
criterion_corr = PearsonCorrelationLoss2("sign")
elif args.loss_type == "PC3_sign":
criterion_corr = PearsonCorrelationLoss3("sign")
else:
raise NotImplementedError()
self.device = torch.device(
"cuda:0" if torch.cuda.is_available() else "cpu")
self.model = model.to(self.device)
self.train_loader, self.validate_loader = make_data_loader(args)
self.optimizer = torch.optim.Adam(self.model.parameters(), lr=args.lr)
self.criterion_ce = criterion_ce.to(self.device)
self.criterion_corr = criterion_corr.to(self.device)
self.best_pred = 0.0
# config saver
self.saver = Saver(args)
self.saver.save_experiment_config()
# tensorboard
self.summary = TensorboardSummary(self.saver.experiment_dir, mean, std)
self.writer = self.summary.create_summary()
# resume training
if args.resume is not None:
if not os.path.isfile(args.resume):
raise RuntimeError("=> no checkpoint found at '{}'".format(
args.resume))
checkpoint = torch.load(args.resume)
args.start_epoch = checkpoint['epoch']
if args.cuda:
self.model.module.load_state_dict(checkpoint['state_dict'])
else:
self.model.load_state_dict(checkpoint['state_dict'])
if not args.ft:
self.optimizer.load_state_dict(checkpoint['optimizer'])
self.best_pred = checkpoint['best_pred']
print("=> loaded checkpoint '{}' (epoch {})".format(
args.resume, checkpoint['epoch']))
def training(self):
pass
def plot(self, logdir, train_loss, valid_loss, **kwargs):
fig = plt.figure(figsize=(10, 8))
plt.plot(range(0, len(train_loss)), train_loss, label='Training Loss')
plt.plot(range(0, len(valid_loss)),
valid_loss,
label='Validation Loss')
for key in kwargs:
plt.plot(range(0, len(kwargs[key])), kwargs[key], label=key)
# find position of lowest validation loss
minposs = valid_loss.index(min(valid_loss))
plt.axvline(minposs,
linestyle='--',
color='r',
label='Early Stopping Checkpoint')
plt.xlabel('epochs')
plt.ylabel('loss')
plt.xlim(0, len(train_loss) + 1) # consistent scale
plt.grid(True)
plt.legend()
plt.tight_layout()
fig.savefig(os.path.join(logdir, 'training_curve.png'),
bbox_inches='tight')
if 'state_dict' in experiment.optimizer:
optimizer.load_state_dict(torch.load(experiment.optimizer.state_dict))
# Logger
if len(experiment.session.log.when) > 0:
logger = SummaryWriter(experiment.session.log.folder)
logger.add_text(
'Experiment',
textwrap.indent(pyaml.dump(experiment, safe=True, sort_dicts=False),
' '), experiment.samples)
else:
logger = None
# Saver
if len(experiment.session.checkpoint.when) > 0:
saver = Saver(experiment.session.checkpoint.folder)
if experiment.epoch == 0:
saver.save_experiment(experiment, suffix=f'e{experiment.epoch:04d}')
else:
saver = None
# endregion
# Datasets and dataloaders
dataset = SolubilityDataset(experiment.session.data.path)
dataloader_kwargs = dict(
num_workers=min(experiment.session.cpus, 1)
if 'cuda' in experiment.session.device else experiment.session.cpus,
pin_memory='cuda' in experiment.session.device,
worker_init_fn=lambda _: np.random.seed(
int(torch.initial_seed()) % (2**32 - 1)),
batch_size=experiment.session.batch_size,
from saver import Saver
env = gym.make(ENV_NAME)
num_actions = env.action_space.n
online_network = DQModel(256, num_actions, str_name="Online")
target_network = DQModel(256, num_actions, str_name="Target")
online_network.compile(optimizer=keras.optimizers.Adam(), loss='mse')
# rendi target_network = primary_network
for t, e in zip(target_network.trainable_variables, online_network.trainable_variables):
t.assign(e)
online_network.compile(optimizer=keras.optimizers.Adam(), loss=tf.keras.losses.Huber())
saver = Saver(ckpt_path=CKPT_PATH, parameters_path=PARAMETERS_PATH)
if LOAD and TRAIN:
try:
episode, total_steps, eps, session = saver.load_parameters()
saver.load_models(online_network, target_network)
except Exception:
episode = 0 # ogni volta che step % SAVE_EACH==0 vengono salvati i weights dei due modelli
total_steps = 0
eps = MAX_EPSILON
session = dt.datetime.now().strftime('%d%m%Y%H%M')
else:
episode = 0 # ogni volta che step % SAVE_EACH==0 vengono salvati i weights dei due modelli
total_steps = 0
eps = MAX_EPSILON
session = dt.datetime.now().strftime('%d%m%Y%H%M')
[6, 21, 31, 36],
[7, 27],
[0, 16, 22, 24, 30],
[19, 29],
[34, 35, 37, 38],
]
task_groups = [task_groups[opt.task]]
num_tasks = len(task_groups)
num_classes = sum([len(elt) for elt in task_groups])
classes = task_groups[0]
# Saving settings
model_dir = os.path.join(opt.checkpoint_path, opt.name)
os.mkdir(model_dir) if not os.path.isdir(model_dir) else None
saver = Saver(model_dir, args=opt)
# Define device, model and optimiser
gpu = utils.check_gpu()
device = torch.device(
"cuda:{}".format(gpu) if torch.cuda.is_available() else "cpu")
model = resnet18(task_groups, trainable_weights=False).to(device)
criterion = nn.BCELoss(reduction='none')
optimizer = optim.Adam(model.parameters(), lr=1e-4)
# Recover weights, if required
if opt.recover:
ckpt_file = os.path.join(model_dir, opt.reco_type + '_weights.pth')
ckpt = torch.load(ckpt_file, map_location=device)
model.load_state_dict(ckpt['model_state_dict'])
epoch = ckpt['iter_nb'] + 1
class Ui_Form(QWidget):
def setupUi(self):
self.setObjectName("Form")
self.resize(1600, 1000)
self.tabWidget = QtWidgets.QTabWidget(self)
self.tabWidget.setGeometry(QtCore.QRect(20, 20, 1151, 131))
self.tabWidget.setObjectName("tabWidget")
self.img = '003.jpg'
self.path = './'
self.photo_shower = PhotoShower()
# self.photo_shower.data.read(self.img)
# self.photo_shower.show(0)
# self.photo_shower.resize(1024, 768)
self.tab_splite = GrayParams(self.photo_shower)#QtWidgets.QWidget()
self.tab_splite.setObjectName("tab_splite")
self.tab_splite.qbt0.clicked.connect(self.changeColorImg)
self.tab_splite.qbt1.clicked.connect(self.changeColorImg)
self.tabWidget.addTab(self.tab_splite, "拆分")
self.tab_open_close = OpenClose(self.photo_shower)#QtWidgets.QWidget()
self.tab_open_close.setObjectName("tab_open_close")
#self.tab_open_close.sld_ksize.valueChanged.connect(self.photo_shower.showProcessedImg)
self.tabWidget.addTab(self.tab_open_close, "开闭")
self.tab_thresh = Thresh(self.photo_shower)
self.tab_thresh.setObjectName('tab_thresh')
self.tabWidget.addTab(self.tab_thresh, "Threshold")
self.tab_filter = Filter(self.photo_shower)
self.tab_filter.setObjectName('filter')
self.tabWidget.addTab(self.tab_filter, '过滤')
self.photo_shower.cur_index = 3
#self.tabWidget.setCurrentIndex(3)
self.tab_beautify = Beautify(self.photo_shower)
self.tab_beautify.setObjectName('beautify')
self.tabWidget.addTab(self.tab_beautify, '美化')
#self.tabWidget.setCurrentIndex(4)
self.tab_saver = Saver(self.photo_shower)
self.tab_saver.setObjectName('saver')
self.tabWidget.addTab(self.tab_saver, '保存')
self.tabWidget.setCurrentIndex(4)
self.horizontalLayoutWidget = QtWidgets.QWidget(self)
self.horizontalLayoutWidget.setGeometry(QtCore.QRect(20, 160, 1151, 811))
self.horizontalLayoutWidget.setObjectName("horizontalLayoutWidget")
self.horizontalLayout = QtWidgets.QHBoxLayout(self.horizontalLayoutWidget)
self.horizontalLayout.setContentsMargins(0, 5, 0, 10)#左，顶，右，底边距
self.horizontalLayout.addWidget(self.photo_shower)
self.horizontalLayout.setObjectName("horizontalLayout")
#self.vlayout
self.img_index = 0#0-3:gray,rgb
self.is_gray = False
self.vlayout_widget = QtWidgets.QWidget(self)
self.vlayout_widget.setGeometry(QtCore.QRect(1100, 0, 351, 900))
self.vlayout_widget.setObjectName("verticalLayoutWidget")
# self.vlayout.addStretch(0)
self.groupBox = QtWidgets.QGroupBox(self.vlayout_widget)
self.groupBox.resize(420, 720)
self.groupBox.setTitle("123")
self.groupBox.setObjectName("groupBox")
self.qbt0 = QtWidgets.QPushButton(self.groupBox)
self.qbt0.setGeometry(QtCore.QRect(10, 20, 150, 30))
self.qbt0.setText('一键处理')
#self.qbt0 = QPushButton('原始图像-->灰度图像')
self.qbt0.clicked.connect(self.processImg)
#self.qbt0.move(0, 0)
self.qbt1 = QtWidgets.QPushButton(self.groupBox)
self.qbt1.setGeometry(QtCore.QRect(10, 60, 150, 30))
self.qbt1.setText('批量处理')
self.qbt1.clicked.connect(self.processImgsTogether)
self.qbt2 = QtWidgets.QPushButton(self.groupBox)
self.qbt2.setGeometry(QtCore.QRect(190, 20, 150, 30))
self.qbt2.setText('转换图像')
self.is_process = True
self.qbt2.clicked.connect(self.restore)
self.qbt3 = QtWidgets.QPushButton(self.groupBox)
self.qbt3.setGeometry(QtCore.QRect(190, 60, 150, 30))
self.qbt3.setText('保存')
self.qbt3.clicked.connect(self.saveImg)
#切换图像
self.label0 = QtWidgets.QLabel(self.groupBox)
self.label0.setGeometry(QtCore.QRect(10, 100, 60, 20))
self.label0.setText('当前图像:')
self.text_img = QtWidgets.QLineEdit(self.groupBox)
self.text_img.setGeometry(QtCore.QRect(70, 100, 210, 20))
self.text_img.setText(self.img)
self.qbt4 = QtWidgets.QPushButton(self.groupBox)
self.qbt4.setGeometry(QtCore.QRect(290, 99, 50, 22))
self.qbt4.setText('...')
self.qbt4.clicked.connect(self.changeImg)
self.slider = QtWidgets.QSlider(Qt.Horizontal, self.groupBox)
self.slider.setGeometry(QtCore.QRect(40, 140, 200, 20))
self.slider.setMinimum(0)
self.slider.setMaximum(200)
self.slider.setValue(100)
self.slider.valueChanged.connect(self.sliderChange)
self.label1 = QtWidgets.QLabel(self.groupBox)
self.label1.setGeometry(QtCore.QRect(10, 140, 40, 20))
self.label1.setText('比例:')
self.text = QtWidgets.QLineEdit(self.groupBox)
self.text.setGeometry(QtCore.QRect(270, 140, 40, 20))
self.text.setText(str(self.slider.value()))
self.text.textChanged.connect(self.kChange)
#输出
self.edit_msg = QtWidgets.QTextEdit(self.groupBox)
self.edit_msg.resize(330, 420)
self.edit_msg.move(10, 170)
self.msg = OutMsg(self.edit_msg)
self.qbt5 = QPushButton('清理', self.groupBox)
self.qbt5.setGeometry(QtCore.QRect(280, 595, 60, 20))
self.qbt5.clicked.connect(self.msg.clear)
self.tab_filter.addOuter(self.msg)
self.vlayout = QtWidgets.QVBoxLayout(self.vlayout_widget)#groupBox)
self.vlayout.setContentsMargins(0, 200, 0, 10) # 左，顶，右，底边距
#self.vlayout.addWidget(self.qbt0)
#self.vlayout.addWidget(self.qbt1)
self.groupBox.setFixedWidth(350)
self.groupBox.setFixedHeight(620)
self.vlayout.addWidget(self.groupBox)
#self.vlayout.addWidget(self.groupBox)
self.vlayout.addStretch()
self.horizontalLayout.addLayout(self.vlayout)
#self.retranslateUi(Form)
#QtCore.QMetaObject.connectSlotsByName(Form)
self.show()
def retranslateUi(self):
_translate = QtCore.QCoreApplication.translate
self.setWindowTitle(_translate("Form", "Form"))
self.tabWidget.setTabText(self.tabWidget.indexOf(self.tab), _translate("Form", "Tab 1"))
self.tabWidget.setTabText(self.tabWidget.indexOf(self.tab_2), _translate("Form", "Tab 2"))
def changeColorImg(self):
index = 0
if not self.tab_splite.is_rgb:
if not self.tab_splite.is_color:
index = 4
else:
index = self.tab_splite.img_index + 1
self.photo_shower.show(index)
def changeRawGray(self):
self.is_gray = not self.is_gray
if self.is_gray:
self.photo_shower.show(4)
self.qbt0.setText('灰度图像-->彩色图像')
else:
self.photo_shower.show(0)
self.qbt0.setText('彩色图像-->灰度图像')
def processImgsTogether(self):
if not os.path.exists('image.txt'):
print("找不到待处理图片文件！")
f = open('image.txt','r')
for line in f.readlines():
img_name = line.replace('\n','')
if self.photo_shower.data.read(img_name):
self.tab_open_close.initData(self.photo_shower)
self.tab_thresh.initData(self.photo_shower)
self.tab_filter.initData(self.photo_shower)
self.processImg()
pos = img_name.rfind('.')
img_color_name = img_name[:pos] + '_color.jpg'
img_black_name = img_name[:pos] + '_black.jpg'
print(img_color_name,'\n', img_black_name)
width = self.photo_shower.data.raw_width
height = self.photo_shower.data.raw_height
img_color = cv2.resize(self.photo_shower.data.img_show, (width, height))
img_color = cv2.cvtColor(img_color, cv2.COLOR_BGR2RGB)
cv2.imwrite(img_color_name, img_color)
img_black = cv2.resize(self.photo_shower.data.img_binary, (width, height))
cv2.imwrite(img_black_name, img_black)
print('读取成功!')
else:
print('读取失败!')
#self.processImg()
# color_list = ['灰', '红', '绿', '蓝']
# self.img_index = (self.img_index + 1) % 4
# if 0 == self.img_index:
# self.photo_shower.show(4)
# else:
# self.photo_shower.show(self.img_index)
# self.qbt1.setText('图像切换:' + color_list[self.img_index] + '-->' + color_list[(self.img_index + 1) % 4])
def restore(self):
if self.is_process:
self.qbt2.setText('原始图像')
self.photo_shower.show(5)
else:
self.qbt2.setText('处理后图像')
self.photo_shower.showProcessedImg()
self.is_process = not self.is_process
def sliderChange(self):
self.text.setText(str(self.slider.value()))
def kChange(self):
self.slider.setValue(int(self.text.text()))
pass
def changeImg(self):
self.img, chose = QtWidgets.QFileDialog.getOpenFileName(self,
"选择图像", self.path,
"ALL(*);;JPG (*.jpg);;PNG (*.png);;BMP(*.bmp)")
#self.path = ''
#print(self.img)
if self.photo_shower.data.read(self.img):
self.photo_shower.show(0)
self.text_img.setText(self.img)
self.tab_splite.initData(self.photo_shower)
self.tab_open_close.initData(self.photo_shower)
self.tab_thresh.initData(self.photo_shower)
self.tab_filter.initData(self.photo_shower)
pos = self.img.rfind('/')
if pos > 0:
self.path = self.img[:pos]
else:
pos = self.img.rfind('\\')
if pos > 0:
self.path = self.img[:pos]
print('读取成功!')
else:
print('读取失败!')
def saveImg(self):
data = self.photo_shower.data
img_name = data.img_name
pos = img_name.rfind('.')
if pos > 0:
img_type = img_name[pos:len(img_name)]
img_type = '.jpg'
img_name = img_name[:pos] + '_after' + img_type
#img_name = img_name[:pos] + img_type
print(data.raw_width, data.raw_height)
img = cv2.resize(data.img_show, (data.raw_width, data.raw_height))
cv2.imwrite(img_name, img)
def processImg(self):
self.tab_splite.changeImg(1)#(self.tab_splite.cur_index)
self.tab_open_close.processImg()
self.tab_thresh.processImgByThresh()
self.tab_filter.filterSize()
self.tab_filter.fufillImg()
self.tab_filter.rectangulazie()
def __init__(self, subj, session_params, cam_params=default_cam_params):
self.params = self.DEFAULT_PARAMS
self.cam_params = cam_params
self.subj = subj
self.name = time.strftime("%Y%m%d_%H%M%S")
self.saver = Saver(self.subj, self.name)
# kwargs
self.params.update(session_params)
# checks on kwargs
assert self.params["min_isi"] > self.params["stim_duration"] # otherwise same-side stims can overlap
if self.params["lick_rule_side"]:
assert self.params[
"lick_rule_any"
] # if must lick correct side in lick phase, then must lick at all in lick phase
# extensions of kwargs
if not isinstance(self.params["lam"], list):
self.params["lam"] = [self.params["lam"]]
if not isinstance(self.params["n_trials"], list):
self.params["n_trials"] = [self.params["n_trials"]]
assert len(self.params["lam"]) == len(self.params["n_trials"])
self.params["subj_name"] = self.subj.name
# self.params['phase_durations'][self.PHASE_STIM] = self.params['stim_phase_intro_duration'] + self.params['stim_phase_duration']
self.cam_params.update(dict(save_name=pjoin(self.subj.subj_dir, self.name)))
if self.params["hints_on"] is True:
self.params["hints_on"] = 1e6
# add them in
self.__dict__.update(self.params)
# hardware
syncobj = multiprocessing.Value("d", 0)
threading.Thread(target=update_sync, args=(syncobj,)).start()
self.cam = PSEye(clock_sync_obj=syncobj, **self.cam_params)
self.cam.start()
self.lr = AnalogReader(
saver=self.saver, lick_thresh=5.0, ports=["ai0", "ai1", "ai5", "ai6"], runtime_ports=[0, 1]
)
sd = self.stim_duration
if self.stim_duration_override:
sd = self.stim_duration_override
self.stimulator = Valve(saver=self.saver, ports=["port0/line0", "port0/line1"], name="stimulator", duration=sd)
self.spout = Valve(
saver=self.saver, ports=["port0/line2", "port0/line3"], name="spout", duration=self.reward_duration
)
self.light = Light(saver=self.saver, port="ao0")
self.speaker = Speaker(saver=self.saver)
# trials init
self.trials = self.generate_trials()
# save session info
self.saver.save_session(self)
# runtime variables
self.session_on = 0
self.session_complete = False
self.session_kill = False
self.trial_on = 0
self.trial_idx = -1
self.valid_trial_idx = 0
self.saving_trial_idx = 0
self.session_runtime = -1
self.trial_runtime = -1
self.trial_outcomes = []
self.trial_corrects = [] # for use in use_trials==False
self.rewards_given = 0
self.paused = 0
self.holding = False
self.percentages = [0.0, 0.0] # L/R
self.side_ns = [0, 0] # L/R
self.bias_correction_percentages = [0.0, 0.0] # L/R
self.perc_valid = 0.0
self.iter_write_begin = now()
def main():
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
load_weights_folder = os.path.expanduser(settings.load_weights_dir)
model_path = os.path.join(load_weights_folder, "model.pth")
model_dict = torch.load(model_path)
# data
datasets_dict = {
"3d60": datasets.ThreeD60,
"panosuncg": datasets.PanoSunCG,
"stanford2d3d": datasets.Stanford2D3D,
"matterport3d": datasets.Matterport3D
}
dataset = datasets_dict[settings.dataset]
fpath = os.path.join(os.path.dirname(__file__), "datasets", "{}_{}.txt")
test_file_list = fpath.format(settings.dataset, "test")
test_dataset = dataset(settings.data_path,
test_file_list,
model_dict['height'],
model_dict['width'],
is_training=False)
test_loader = DataLoader(test_dataset,
settings.batch_size,
False,
num_workers=settings.num_workers,
pin_memory=True,
drop_last=False)
num_test_samples = len(test_dataset)
num_steps = num_test_samples // settings.batch_size
print("Num. of test samples:", num_test_samples, "Num. of steps:",
num_steps, "\n")
# network
Net_dict = {"UniFuse": UniFuse, "Equi": Equi}
Net = Net_dict[model_dict['net']]
model = Net(model_dict['layers'],
model_dict['height'],
model_dict['width'],
max_depth=test_dataset.max_depth_meters,
fusion_type=model_dict['fusion'],
se_in_fusion=model_dict['se_in_fusion'])
model.to(device)
model_state_dict = model.state_dict()
model.load_state_dict(
{k: v
for k, v in model_dict.items() if k in model_state_dict})
model.eval()
evaluator = Evaluator(settings.median_align)
evaluator.reset_eval_metrics()
saver = Saver(load_weights_folder)
pbar = tqdm.tqdm(test_loader)
pbar.set_description("Testing")
with torch.no_grad():
for batch_idx, inputs in enumerate(pbar):
equi_inputs = inputs["normalized_rgb"].to(device)
cube_inputs = inputs["normalized_cube_rgb"].to(device)
outputs = model(equi_inputs, cube_inputs)
pred_depth = outputs["pred_depth"].detach().cpu()
gt_depth = inputs["gt_depth"]
mask = inputs["val_mask"]
for i in range(gt_depth.shape[0]):
evaluator.compute_eval_metrics(gt_depth[i:i + 1],
pred_depth[i:i + 1],
mask[i:i + 1])
if settings.save_samples:
saver.save_samples(inputs["rgb"], gt_depth, pred_depth, mask)
evaluator.print(load_weights_folder)
class Session(object):
def __init__(self, session_params, ax_interactive=None):
# incorporate kwargs
self.params = session_params
self.__dict__.update(self.params)
self.verify_params()
# sync
self.sync_flag = multiprocessing.Value('b', False)
self.sync_to_save = multiprocessing.Queue()
# saver
self.saver = Saver(self.subj, self.name, self, sync_flag=self.sync_flag)
# hardware
self.cam = PSEye(sync_flag=self.sync_flag, **self.cam_params)
self.ar = AnalogReader(saver_obj_buffer=self.saver.buf, sync_flag=self.sync_flag, **self.ar_params)
# communication
self.ni = NI845x(i2c_on=self.imaging)
# interactivity
self.ax_interactive = ax_interactive
# runtime variables
self.notes = {}
self.mask_idx = -1 #for reselecting mask
self.session_on = 0
self.on = False
self.session_complete = False
self.session_kill = False
self.trial_flag = False
self.trial_on = 0
self.trial_off = 0
self.trial_idx = -1
self.stim_cycle_idx = 0
self.paused = False
self.deliver_override = False
self.roi_pts = None
self.eyelid_buffer = np.zeros(self.eyelid_buffer_size)-1
self.eyelid_buffer_ts = np.zeros(self.eyelid_buffer_size)-1
self.past_flag = False
# sync
self.sync_flag.value = True #trigger all processes to get time
self.sync_val = now() #get this process's time
procs = dict(saver=self.saver, cam=self.cam.pseye, ar=self.ar)
sync_vals = {o:procs[o].sync_val.value for o in procs} #collect all process times
sync_vals['session'] = self.sync_val
self.sync_to_save.put(sync_vals)
# more runtime, anything that must occur after sync
_,self.im = self.cam.get()
@property
def session_runtime(self):
if self.session_on != 0:
return now()-self.session_on
else:
return -1
@property
def trial_runtime(self):
if self.trial_on != False:
return now()-self.trial_on
else:
return -1
def name_as_str(self):
return self.name.strftime('%Y%m%d%H%M%S')
def verify_params(self):
if self.name is None:
self.name = pd.datetime.now()
self.cam_params.update(dict(save_name=pjoin(self.subj.subj_dir, self.name_as_str()+'_cams.h5')))
def pause(self, val):
self.paused = val
if self.imaging:
if val == True:
self.stop_acq()
elif val == False:
self.start_acq()
def update_licked(self):
l = self.ar.licked
def start_acq(self):
if self.imaging:
self.ni.write_dio(LINE_SI_ON, 1)
self.ni.write_dio(LINE_SI_ON, 0)
def stop_acq(self):
if self.imaging:
self.ni.write_dio(LINE_SI_OFF, 1)
self.ni.write_dio(LINE_SI_OFF, 0)
def wait(self, dur, t0=None):
if t0 is None:
t0 = now()
while now()-t0 < dur:
pass
def next_stim_type(self, inc=True):
st = self.cycle[self.stim_cycle_idx]
if inc:
self.stim_cycle_idx += 1
if self.stim_cycle_idx == len(self.cycle):
self.stim_cycle_idx = 0
return st
@property
def current_stim_state(self):
return STIM_TYPES[self.cycle[self.stim_cycle_idx]]
def deliver_trial(self):
while self.on:
if self.trial_flag:
# prepare trial
self.trial_idx += 1
self.trial_on = now()
self.cam.set_flush(False)
kind = self.next_stim_type()
# deilver trial
self.wait(self.intro)
cs_time,us_time = self.send_stim(kind)
# replay
self.wait(self.display_lag)
self.past_flag = [cs_time[1], us_time[1]]
# finish trial
self.wait(self.trial_duration, t0=self.trial_on)
self.trial_off = now()
# save trial info
self.cam.set_flush(True)
trial_dict = dict(\
start = self.trial_on,\
end = self.trial_off,\
cs_ts0 = cs_time[0],\
cs_ts1 = cs_time[1],\
us_ts0 = us_time[0],\
us_ts1 = us_time[1],\
kind = kind,\
idx = self.trial_idx,\
)
self.saver.write('trials',trial_dict)
self.trial_flag = False
self.trial_on = False
def dummy_puff(self):
self.ni.write_dio(LINE_US, 1)
self.wait(self.us_dur)
self.ni.write_dio(LINE_US, 0)
def dummy_light(self, state):
self.ni.write_dio(LINE_CS, state)
def send_stim(self, kind):
if kind == CS:
t = (now(), now2())
self.ni.write_i2c('CS_ON')
self.ni.write_dio(LINE_CS, 1)
self.wait(self.cs_dur)
self.ni.write_i2c('CS_OFF')
self.ni.write_dio(LINE_CS, 0)
stim_time = [t,(-1,-1)]
elif kind == US:
self.wait(self.cs_dur) # for trial continuity
t = (now(), now2())
self.ni.write_i2c('US_ON')
self.ni.write_dio(LINE_US, 1)
self.wait(self.us_dur)
self.ni.write_i2c('US_OFF')
self.ni.write_dio(LINE_US, 0)
stim_time = [(-1,-1),t]
elif kind == CSUS:
t_cs = (now(), now2())
self.ni.write_i2c('CS_ON')
self.ni.write_dio(LINE_CS, 1)
self.wait(self.csus_gap)
t_us = (now(), now2())
self.ni.write_i2c('US_ON')
self.ni.write_dio(LINE_US, 1)
self.wait(self.us_dur) # assumes US ends before CS does
self.ni.write_i2c('US_OFF')
self.ni.write_dio(LINE_US, 0)
self.wait(self.cs_dur, t0=t_cs[0])
self.ni.write_i2c('CS_OFF')
self.ni.write_dio(LINE_CS, 0)
stim_time = [t_cs,t_us]
return stim_time
def acquire_mask(self):
x,y = self.cam.resolution[0]
if self.roi_pts is None:
self.roi_pts = [[0,0],[x,0],[x,y],[0,y]]
logging.warning('No ROI found, using default')
self.mask_idx += 1
pts_eye = np.array(self.roi_pts, dtype=np.int32)
mask_eye = np.zeros([y,x], dtype=np.int32)
cv2.fillConvexPoly(mask_eye, pts_eye, (1,1,1), lineType=cv2.LINE_AA)
self.mask = mask_eye
self.mask_flat = self.mask.reshape((1,-1))
self.saver.write('mask{}'.format(self.mask_idx), self.mask)
logging.info('New mask set.')
def run(self):
try:
self.acquire_mask()
self.session_on = now()
self.on = True
self.ar.begin_saving()
self.cam.begin_saving()
self.cam.set_flush(True)
self.start_acq()
# main loop
threading.Thread(target=self.deliver_trial).start()
threading.Thread(target=self.update_eyelid).start()
while True:
if self.trial_on or self.paused:
continue
if self.session_kill:
break
moving = self.determine_motion()
eyelid = self.determine_eyelid()
if self.deliver_override or ((now()-self.trial_off>self.min_iti) and (not moving) and (eyelid)):
self.trial_flag = True
self.deliver_override = False
self.end()
except:
logging.error('Session has encountered an error!')
raise
def determine_eyelid(self):
return np.mean(self.eyelid_buffer[-self.eyelid_window:]) < self.eyelid_thresh
def update_eyelid(self):
while self.on:
imts,im = self.cam.get()
if im is None:
continue
self.im = im
roi_data = self.extract(self.im)
self.eyelid_buffer = np.roll(self.eyelid_buffer, -1)
self.eyelid_buffer_ts = np.roll(self.eyelid_buffer_ts, -1)
self.eyelid_buffer[-1] = roi_data
self.eyelid_buffer_ts[-1] = imts
def extract(self, fr):
if fr is None:
return 0
flat = fr.reshape((1,-1)).T
dp = (self.mask_flat.dot(flat)).T
return np.squeeze(dp/self.mask_flat.sum(axis=-1))
def determine_motion(self):
return self.ar.moving
def end(self):
self.on = False
self.stop_acq()
to_end = [self.ar, self.cam]
if self.imaging:
to_end.append(self.ni)
for te in to_end:
te.end()
time.sleep(0.100)
self.saver.end(notes=self.notes)
self.session_on = False
def get_code(self):
py_files = [pjoin(d,f) for d,_,fs in os.walk(os.getcwd()) for f in fs if f.endswith('.py') and not f.startswith('__')]
code = {}
for pf in py_files:
with open(pf, 'r') as f:
code[pf] = f.read()
return json.dumps(code)
def main(_=None):
with tf.Session(config=tf.ConfigProto(allow_soft_placement=True)) as sess:
d = os.path.dirname(args.log_file)
if not os.path.exists(d):
os.makedirs(d)
if not args.continue_training:
with open(args.log_file, 'w') as f:
f.write('')
logging.basicConfig(filename=args.log_file, level=logging.DEBUG)
logging.getLogger().addHandler(logging.StreamHandler())
game_handler = GameStateHandler(
args.rom_directory + args.rom_name,
random_seed=args.random_seed,
frame_skip=args.frame_skip,
use_sdl=args.use_sdl,
repeat_action_probability=args.repeat_action_probability,
minimum_actions=args.minimum_action_set,
test_mode=args.test_mode,
image_processing=lambda x: crop_and_resize(x, args.image_height, args.image_width, args.cut_top))
num_actions = game_handler.num_actions
if args.optimizer == 'rmsprop':
optimizer = tf.train.RMSPropOptimizer(
learning_rate=args.learning_rate,
decay=args.decay,
momentum=0.0,
epsilon=args.rmsprop_epsilon)
if not args.multi_gpu:
if args.double_dqn:
net = qnetwork.DualDeepQNetwork(args.image_height, args.image_width, sess, num_actions,
args.state_frames, args.discount_factor, args.target_net_refresh_rate,
net_type=args.net_type, optimizer=optimizer)
else:
net = qnetwork.DeepQNetwork(args.image_height, args.image_width, sess, num_actions, args.state_frames,
args.discount_factor, net_type=args.net_type, optimizer=optimizer)
else:
net = multi_gpu_qnetwork.MultiGPUDualDeepQNetwork(args.image_height, args.image_width, sess, num_actions,
args.state_frames, args.discount_factor,
optimizer=optimizer, gpus=[0, 1, 2, 3])
saver = Saver(sess, args.data_dir, args.continue_training)
if saver.replay_memory_found():
replay_memory = saver.get_replay_memory()
else:
if args.test_mode:
logging.error('NO SAVED NETWORKS IN TEST MODE!!!')
replay_memory = ReplayMemoryManager(args.image_height, args.image_width, args.state_frames,
args.replay_memory_size, reward_clip_min=args.reward_clip_min,
reward_clip_max=args.reward_clip_max)
# todo: add parameters to handle monitor
monitor = Monitoring(log_train_step_every=100, smooth_episode_scores_over=50)
agent = Agent(
game_handler=game_handler,
qnetwork=net,
replay_memory=replay_memory,
saver=saver,
monitor=monitor,
train_freq=args.train_freq,
test_mode=args.test_mode,
batch_size=args.batch_size,
save_every_x_episodes=args.saving_freq)
sess.run(tf.initialize_all_variables())
saver.restore(args.data_dir)
start_epsilon = max(args.final_epsilon,
args.start_epsilon - saver.get_start_frame() * (args.start_epsilon - args.final_epsilon) / args.exploration_duration)
exploring_duration = max(args.exploration_duration - saver.get_start_frame(), 1)
if args.test_mode:
agent.populate_replay_memory(args.state_frames, force_early_stop=True)
agent.play_in_test_mode(args.epsilon_in_test_mode)
else:
agent.populate_replay_memory(args.min_replay_memory)
agent.play(train_steps_limit=args.number_of_train_steps, start_eps=start_epsilon,
final_eps=args.final_epsilon, exploring_duration=exploring_duration)
if trash:
[trh.extract() for trh in trash]
except IndexError:
trash = body.findAll(tag)
if trash:
[trh.extract() for trh in trash]
comments = body.findAll(text=lambda text:isinstance(text, Comment))
[comment.extract() for comment in comments]
return body
# получаем настройки
def getSetings(self):
if not 'http://' in self.url: self.url = 'http://' + self.url
n = self.url.replace('www.', '').split('/')[2]
try:
self.settings = SITES[n]
except KeyError:
self.settings = SITES['default']
# разбиваем id/class:name по двоеточию
def indent(self, txt):
return txt.split(':')
#тест на 4х сайтах по 10 урлов с каждого
if __name__ == '__main__':
from saver import Saver
for line in open('testLink.txt'):
line = line.rstrip()
test = Parser(line)
f = Saver(line)
f.saveFile(test.result())
# -*- coding: utf-8 -*-
from parser import Parser
from saver import Saver
import sys
try:
url = sys.argv[1]
except IndexError:
print 'Не передан URL'
sys.exit()
obj = Parser(url)
text = obj.result()
f = Saver(url)
f.saveFile(text)
def __save__(self):
saver = Saver()
for obj in self.mp:
saver.save_in_file(obj, self.mp[obj])
def __init__(self, json_data, dir):
Saver.__init__(self, json_data, dir, QzonePath.BLOG)
self._filename = "category_info.json"
self._cate_info = self.json_data["data"]["cateInfo"]
def main():
# parse options
parser = TrainOptions()
opts = parser.parse()
# daita loader
print('\n--- load dataset ---')
dataset = dataset_multi(opts)
train_loader = torch.utils.data.DataLoader(dataset,
batch_size=opts.batch_size,
shuffle=True,
num_workers=opts.nThreads)
# model
print('\n--- load model ---')
model = SAVI2I(opts)
model.setgpu(opts.gpu)
if opts.resume is None:
model.initialize()
ep0 = -1
total_it = 0
else:
ep0, total_it = model.resume(opts.resume)
model.set_scheduler(opts, last_ep=ep0)
ep0 += 1
print('start the training at epoch %d' % (ep0))
# saver for display and output
saver = Saver(opts)
# train
print('\n--- train ---')
max_it = 1000000
for ep in range(ep0, opts.n_ep):
for it, (images, c_org, c_org_mask,
c_org_id) in enumerate(train_loader):
# input data
images = torch.cat(images, dim=0)
images = images.cuda(opts.gpu).detach()
c_org = torch.cat(c_org, dim=0)
c_org = c_org.cuda(opts.gpu).detach()
c_org_mask = torch.cat(c_org_mask, dim=0)
c_org_mask = c_org_mask.cuda(opts.gpu).detach()
c_org_id = torch.cat(c_org_id, dim=0)
c_org_id = c_org_id.cuda(opts.gpu).detach()
# update model
if (it + 1) % opts.d_iter != 0 and it < len(train_loader) - 2:
model.update_D_content(images, c_org)
continue
else:
model.update_D(images, c_org, c_org_mask, c_org_id)
model.update_EFG()
print('total_it: %d (ep %d, it %d), lr %08f' %
(total_it, ep, it, model.gen_opt.param_groups[0]['lr']))
total_it += 1
if total_it >= max_it:
saver.write_img(-1, model)
saver.write_model(-1, max_it, model)
break
# decay learning rate
if opts.n_ep_decay > -1:
model.update_lr()
# save result image
saver.write_img(ep, model)
# Save network weights
saver.write_model(ep, total_it, model)
return
print("""
Runnung with arguments:
Input file: %s
Use proxy: %s
Requests per minute: %0.2f (if proxy in use, then for a proxy)
Max. proxy usage: %d (if less then 2 proxies in use, ignored)
Download files: %s
""" % (INPUT_FILENAME, USE_PROXY, REQUESTS_PER_MINUTE,
MAX_PROXY_USAGE_NUMBER, DO_DOWNLOAD))
if USE_PROXY:
proxy_list.load_from_file(PROXIES_FILENAME)
saver = Saver(autoflush=False)
# mapping sites to their respective classes
site_mapping = [
#((types,), processing function)
[("GBA", "GBP"),
SiteGBProcessor(saver, REQUESTS_PER_MINUTE, MAX_PROXY_USAGE_NUMBER,
do_download=DO_DOWNLOAD)],
[("FIA", "FIP"),
SiteFIProcessor(saver, REQUESTS_PER_MINUTE, MAX_PROXY_USAGE_NUMBER,
do_download=DO_DOWNLOAD, proxy_list=[])],
[("WO"),
SiteWOProcessor(saver, REQUESTS_PER_MINUTE, MAX_PROXY_USAGE_NUMBER,
do_download=DO_DOWNLOAD)],
[("EPA"),
from preprocessor import Preprocessor
from posprocessor import Posprocessor
from analyser import Analyser
from listener import Listener
from saver import Saver
import pandas as pd
import json
newSaver = Saver()
def start_to_listen():
with open('credentials/twitter-api.json') as json_file:
#Open credentials and insert api keys
twitter_apiKeys = json.load(json_file)
print("\n ===== Hey there, I'll start to listen to tweets about Joker ===== \n")
print(" Press ctrl+C when you want me to stop!! ")
mylistener = Listener(twitter_apiKeys)
mylistener.set_authentication()
mylistener.start_listening()
def start_to_preprocess():
pp = Preprocessor()
data = pd.DataFrame(list(newSaver.get_collection('realTimeTweets')))
pp.clean_frame(data)
##use this if you want to create a csv file to visualize
# with open('tmpcsv.csv', 'w') as tmpcsv:
# tmpcsv.writelines(data.to_csv())
plt.tight_layout(pad=2.0, w_pad=0.0, h_pad=0.0)
plt.savefig(fname, dpi=150)
plt.close('all')
def preprocess(x1, x2, y):
"""Preprocess training data."""
return (x1.astype('float32') / 255,
x2.astype('float32') / 255,
y.astype('float32'))
if __name__ == '__main__':
restore_folder = sys.argv[1]
saver = Saver(restore_folder)
ckpt_info = saver.get_ckpt_info()
model_opt = ckpt_info['model_opt']
data_opt = ckpt_info['data_opt']
ckpt_fname = ckpt_info['ckpt_fname']
step = ckpt_info['step']
model_id = ckpt_info['model_id']
log.info('Building model')
m = get_model(model_opt)
log.info('Loading dataset')
dataset = get_dataset(data_opt)
sess = tf.Session()
saver.restore(sess, ckpt_fname)
def __init__(self, json_data, directory):
Saver.__init__(self, json_data, directory, QzonePath.PHOTO)
def setupUi(self):
self.setObjectName("Form")
self.resize(1600, 1000)
self.tabWidget = QtWidgets.QTabWidget(self)
self.tabWidget.setGeometry(QtCore.QRect(20, 20, 1151, 131))
self.tabWidget.setObjectName("tabWidget")
self.img = '003.jpg'
self.path = './'
self.photo_shower = PhotoShower()
# self.photo_shower.data.read(self.img)
# self.photo_shower.show(0)
# self.photo_shower.resize(1024, 768)
self.tab_splite = GrayParams(self.photo_shower)#QtWidgets.QWidget()
self.tab_splite.setObjectName("tab_splite")
self.tab_splite.qbt0.clicked.connect(self.changeColorImg)
self.tab_splite.qbt1.clicked.connect(self.changeColorImg)
self.tabWidget.addTab(self.tab_splite, "拆分")
self.tab_open_close = OpenClose(self.photo_shower)#QtWidgets.QWidget()
self.tab_open_close.setObjectName("tab_open_close")
#self.tab_open_close.sld_ksize.valueChanged.connect(self.photo_shower.showProcessedImg)
self.tabWidget.addTab(self.tab_open_close, "开闭")
self.tab_thresh = Thresh(self.photo_shower)
self.tab_thresh.setObjectName('tab_thresh')
self.tabWidget.addTab(self.tab_thresh, "Threshold")
self.tab_filter = Filter(self.photo_shower)
self.tab_filter.setObjectName('filter')
self.tabWidget.addTab(self.tab_filter, '过滤')
self.photo_shower.cur_index = 3
#self.tabWidget.setCurrentIndex(3)
self.tab_beautify = Beautify(self.photo_shower)
self.tab_beautify.setObjectName('beautify')
self.tabWidget.addTab(self.tab_beautify, '美化')
#self.tabWidget.setCurrentIndex(4)
self.tab_saver = Saver(self.photo_shower)
self.tab_saver.setObjectName('saver')
self.tabWidget.addTab(self.tab_saver, '保存')
self.tabWidget.setCurrentIndex(4)
self.horizontalLayoutWidget = QtWidgets.QWidget(self)
self.horizontalLayoutWidget.setGeometry(QtCore.QRect(20, 160, 1151, 811))
self.horizontalLayoutWidget.setObjectName("horizontalLayoutWidget")
self.horizontalLayout = QtWidgets.QHBoxLayout(self.horizontalLayoutWidget)
self.horizontalLayout.setContentsMargins(0, 5, 0, 10)#左，顶，右，底边距
self.horizontalLayout.addWidget(self.photo_shower)
self.horizontalLayout.setObjectName("horizontalLayout")
#self.vlayout
self.img_index = 0#0-3:gray,rgb
self.is_gray = False
self.vlayout_widget = QtWidgets.QWidget(self)
self.vlayout_widget.setGeometry(QtCore.QRect(1100, 0, 351, 900))
self.vlayout_widget.setObjectName("verticalLayoutWidget")
# self.vlayout.addStretch(0)
self.groupBox = QtWidgets.QGroupBox(self.vlayout_widget)
self.groupBox.resize(420, 720)
self.groupBox.setTitle("123")
self.groupBox.setObjectName("groupBox")
self.qbt0 = QtWidgets.QPushButton(self.groupBox)
self.qbt0.setGeometry(QtCore.QRect(10, 20, 150, 30))
self.qbt0.setText('一键处理')
#self.qbt0 = QPushButton('原始图像-->灰度图像')
self.qbt0.clicked.connect(self.processImg)
#self.qbt0.move(0, 0)
self.qbt1 = QtWidgets.QPushButton(self.groupBox)
self.qbt1.setGeometry(QtCore.QRect(10, 60, 150, 30))
self.qbt1.setText('批量处理')
self.qbt1.clicked.connect(self.processImgsTogether)
self.qbt2 = QtWidgets.QPushButton(self.groupBox)
self.qbt2.setGeometry(QtCore.QRect(190, 20, 150, 30))
self.qbt2.setText('转换图像')
self.is_process = True
self.qbt2.clicked.connect(self.restore)
self.qbt3 = QtWidgets.QPushButton(self.groupBox)
self.qbt3.setGeometry(QtCore.QRect(190, 60, 150, 30))
self.qbt3.setText('保存')
self.qbt3.clicked.connect(self.saveImg)
#切换图像
self.label0 = QtWidgets.QLabel(self.groupBox)
self.label0.setGeometry(QtCore.QRect(10, 100, 60, 20))
self.label0.setText('当前图像:')
self.text_img = QtWidgets.QLineEdit(self.groupBox)
self.text_img.setGeometry(QtCore.QRect(70, 100, 210, 20))
self.text_img.setText(self.img)
self.qbt4 = QtWidgets.QPushButton(self.groupBox)
self.qbt4.setGeometry(QtCore.QRect(290, 99, 50, 22))
self.qbt4.setText('...')
self.qbt4.clicked.connect(self.changeImg)
self.slider = QtWidgets.QSlider(Qt.Horizontal, self.groupBox)
self.slider.setGeometry(QtCore.QRect(40, 140, 200, 20))
self.slider.setMinimum(0)
self.slider.setMaximum(200)
self.slider.setValue(100)
self.slider.valueChanged.connect(self.sliderChange)
self.label1 = QtWidgets.QLabel(self.groupBox)
self.label1.setGeometry(QtCore.QRect(10, 140, 40, 20))
self.label1.setText('比例:')
self.text = QtWidgets.QLineEdit(self.groupBox)
self.text.setGeometry(QtCore.QRect(270, 140, 40, 20))
self.text.setText(str(self.slider.value()))
self.text.textChanged.connect(self.kChange)
#输出
self.edit_msg = QtWidgets.QTextEdit(self.groupBox)
self.edit_msg.resize(330, 420)
self.edit_msg.move(10, 170)
self.msg = OutMsg(self.edit_msg)
self.qbt5 = QPushButton('清理', self.groupBox)
self.qbt5.setGeometry(QtCore.QRect(280, 595, 60, 20))
self.qbt5.clicked.connect(self.msg.clear)
self.tab_filter.addOuter(self.msg)
self.vlayout = QtWidgets.QVBoxLayout(self.vlayout_widget)#groupBox)
self.vlayout.setContentsMargins(0, 200, 0, 10) # 左，顶，右，底边距
#self.vlayout.addWidget(self.qbt0)
#self.vlayout.addWidget(self.qbt1)
self.groupBox.setFixedWidth(350)
self.groupBox.setFixedHeight(620)
self.vlayout.addWidget(self.groupBox)
#self.vlayout.addWidget(self.groupBox)
self.vlayout.addStretch()
self.horizontalLayout.addLayout(self.vlayout)
#self.retranslateUi(Form)
#QtCore.QMetaObject.connectSlotsByName(Form)
self.show()
from validators import DateValidator, resolution_validation
from saver import Saver
if __name__ == "__main__":
WARNING = ('Input parametrs shuold be like\n\'python3 '
'main.py 05 2017 1920x1080\'.')
if len(sys.argv) == 4:
month, year, resolution = sys.argv[1:4]
validator = DateValidator()
if (validator.date_validation(month, year)
and resolution_validation(resolution)):
page_getter = PageGetter()
page = page_getter.get_page(month, year)
page_parser = PageParser(page)
name_link_dict = page_parser.find_all_wallpapers(resolution)
if len(name_link_dict) < 1:
print('You picked a wrong resolution.')
base_dir = os.path.dirname(__file__)
wall_dir = os.path.dirname(base_dir)
path = os.path.join(wall_dir, 'Wallpapers')
if not os.path.isdir(path):
os.mkdir(path)
saver = Saver(path)
saver.save_wallpapers(name_link_dict)
else:
print(WARNING)
else:
print(WARNING)
if load_process:
if user != '' and password != '':
pass
else:
print(
'The user (-u) and the password (-p) options should be specified for proper loading (-l)'
)
pred = predict(input_list)
results = []
for predict, input_string in zip(pred, input_list):
if validate(predict):
result = list(zip(predict, input_string.split()))
results.append(result)
else:
print('Invalid string:')
print(input_string)
if load_process or export_process:
db = DBWrapper(user, password)
if load_process:
publications = [compose_publication(result) for result in results]
db.add(publications)
if export_process:
data = db.get()
exporter = Saver()
exporter.save(data, "xlsx", export_file_name)
else:
for result in results:
print(group_publication(result))
def train(autoencoder_config_path, probclass_config_path,
restore_manager: RestoreManager, log_dir_root, datasets: Datasets,
train_flags: TrainFlags, ckpt_interval_hours: float,
description: str):
ae_config, ae_config_rel_path = config_parser.parse(
autoencoder_config_path)
pc_config, pc_config_rel_path = config_parser.parse(probclass_config_path)
print_configs(('ae_config', ae_config), ('pc_config', pc_config))
continue_in_ckpt_dir = restore_manager and restore_manager.continue_in_ckpt_dir
if continue_in_ckpt_dir:
logdir = restore_manager.log_dir
else:
logdir = logdir_helpers.create_unique_log_dir(
[ae_config_rel_path, pc_config_rel_path],
log_dir_root,
restore_dir=restore_manager.ckpt_dir if restore_manager else None)
print(_LOG_DIR_FORMAT.format(logdir))
if description:
_write_to_sheets(logdir_helpers.log_date_from_log_dir(logdir),
ae_config_rel_path,
pc_config_rel_path,
description,
git_ref=_get_git_ref(),
log_dir_root=log_dir_root,
is_continue=continue_in_ckpt_dir)
ae_cls = autoencoder.get_network_cls(ae_config)
pc_cls = probclass.get_network_cls(pc_config)
# Instantiate autoencoder and probability classifier
ae = ae_cls(ae_config)
pc = pc_cls(pc_config, num_centers=ae_config.num_centers)
# train ---
ip_train = inputpipeline.InputPipeline(
inputpipeline.get_dataset(datasets.train),
ae_config.crop_size,
batch_size=ae_config.batch_size,
shuffle=False,
num_preprocess_threads=NUM_PREPROCESS_THREADS,
num_crops_per_img=NUM_CROPS_PER_IMG)
x_train = ip_train.get_batch()
enc_out_train = ae.encode(
x_train, is_training=True) # qbar is masked by the heatmap
x_out_train = ae.decode(enc_out_train.qbar, is_training=True)
# stop_gradient is beneficial for training. it prevents multiple gradients flowing into the heatmap.
pc_in = tf.stop_gradient(enc_out_train.qbar)
bc_train = pc.bitcost(pc_in,
enc_out_train.symbols,
is_training=True,
pad_value=pc.auto_pad_value(ae))
bpp_train = bits.bitcost_to_bpp(bc_train, x_train)
d_train = Distortions(ae_config, x_train, x_out_train, is_training=True)
# summing over channel dimension gives 2D heatmap
heatmap2D = (tf.reduce_sum(enc_out_train.heatmap, 1)
if enc_out_train.heatmap is not None else None)
# loss ---
total_loss, H_real, pc_comps, ae_comps = get_loss(ae_config, ae, pc,
d_train.d_loss_scaled,
bc_train,
enc_out_train.heatmap)
train_op = get_train_op(ae_config, pc_config, ip_train, pc.variables(),
total_loss)
# test ---
with tf.name_scope('test'):
ip_test = inputpipeline.InputPipeline(
inputpipeline.get_dataset(datasets.test),
ae_config.crop_size,
batch_size=ae_config.batch_size,
num_preprocess_threads=NUM_PREPROCESS_THREADS,
num_crops_per_img=1,
big_queues=False,
shuffle=False)
x_test = ip_test.get_batch()
enc_out_test = ae.encode(x_test, is_training=False)
x_out_test = ae.decode(enc_out_test.qhard, is_training=False)
bc_test = pc.bitcost(enc_out_test.qhard,
enc_out_test.symbols,
is_training=False,
pad_value=pc.auto_pad_value(ae))
bpp_test = bits.bitcost_to_bpp(bc_test, x_test)
d_test = Distortions(ae_config, x_test, x_out_test, is_training=False)
try: # Try to get codec distnace for current dataset
codec_distance_ms_ssim = CodecDistance(datasets.codec_distance,
codec='bpg',
metric='ms-ssim')
get_distance = functools_ext.catcher(ValueError,
handler=functools_ext.const(
np.nan),
f=codec_distance_ms_ssim.distance)
get_distance = functools_ext.compose(np.float32,
get_distance) # cast to float32
d_BPG_test = tf.py_func(get_distance, [bpp_test, d_test.ms_ssim],
tf.float32,
stateful=False,
name='d_BPG')
d_BPG_test.set_shape(())
except CodecDistanceReadException as e:
print('Cannot compute CodecDistance: {}'.format(e))
d_BPG_test = tf.constant(np.nan, shape=(), name='ConstNaN')
# ---
train_logger = Logger()
test_logger = Logger()
distortion_name = ae_config.distortion_to_minimize
train_logger.add_summaries(d_train.summaries_with_prefix('train'))
# Visualize components of losses
train_logger.add_summaries([
tf.summary.scalar('train/PC_loss/{}'.format(name), comp)
for name, comp in pc_comps
])
train_logger.add_summaries([
tf.summary.scalar('train/AE_loss/{}'.format(name), comp)
for name, comp in ae_comps
])
train_logger.add_summaries([tf.summary.scalar('train/bpp', bpp_train)])
train_logger.add_console_tensor('loss={:.3f}', total_loss)
train_logger.add_console_tensor('ms_ssim={:.3f}', d_train.ms_ssim)
train_logger.add_console_tensor('bpp={:.3f}', bpp_train)
train_logger.add_console_tensor('H_real={:.3f}', H_real)
test_logger.add_summaries(d_test.summaries_with_prefix('test'))
test_logger.add_summaries([
tf.summary.scalar('test/bpp', bpp_test),
tf.summary.scalar('test/distance_BPG_MS-SSIM', d_BPG_test),
tf.summary.image('test/x_in',
prep_for_image_summary(x_test, n=3, name='x_in')),
tf.summary.image('test/x_out',
prep_for_image_summary(x_out_test, n=3, name='x_out'))
])
if heatmap2D is not None:
test_logger.add_summaries([
tf.summary.image(
'test/hm',
prep_for_grayscale_image_summary(heatmap2D,
n=3,
autoscale=True,
name='hm'))
])
test_logger.add_console_tensor('ms_ssim={:.3f}', d_test.ms_ssim)
test_logger.add_console_tensor('bpp={:.3f}', bpp_test)
test_logger.add_summaries([
tf.summary.histogram('centers', ae.get_centers_variable()),
tf.summary.histogram(
'test/qbar', enc_out_test.qbar[:ae_config.batch_size // 2, ...])
])
test_logger.add_console_tensor('d_BPG={:.6f}', d_BPG_test)
test_logger.add_console_tensor(Logger.Numpy1DFormatter('centers={}'),
ae.get_centers_variable())
print('Starting session and queues...')
with tf_helpers.start_queues_in_sess(
init_vars=restore_manager is None) as (sess, coord):
train_logger.finalize_with_sess(sess)
test_logger.finalize_with_sess(sess)
if restore_manager:
restore_manager.restore(sess)
saver = Saver(Saver.ckpt_dir_for_log_dir(logdir),
max_to_keep=1,
keep_checkpoint_every_n_hours=ckpt_interval_hours)
train_loop(ae_config,
sess,
coord,
train_op,
train_logger,
test_logger,
train_flags,
logdir,
saver,
is_restored=restore_manager is not None)
def __init__(self, debug=False):
self.debug = debug
self.saver = Saver()
self.last_tweets = self.maybe_load(FileNames.last_tweets)
self.giphy_keys = self.maybe_load(FileNames.giphy_keys)
def __init__(self, json_data, begin, end, directory):
Saver.__init__(self, json_data, directory, QzonePath.MSG_BOARD)
self._filename = "msg_board_%05d-%05d.json" % (begin, end - 1)
def main():
# parse options
parser = TrainOptions()
opts = parser.parse()
# data loader
print('\n--- load dataset ---')
if opts.multi_modal:
dataset = dataset_unpair_multi(opts)
else:
dataset = dataset_unpair(opts)
train_loader = torch.utils.data.DataLoader(dataset,
batch_size=opts.batch_size,
shuffle=True,
num_workers=opts.nThreads)
# model
print('\n--- load model ---')
model = DRIT(opts)
model.setgpu(opts.gpu)
if opts.resume is None:
model.initialize()
ep0 = -1
total_it = 0
else:
ep0, total_it = model.resume(opts.resume)
model.set_scheduler(opts, last_ep=ep0)
ep0 += 1
print('start the training at epoch %d' % (ep0))
# saver for display and output
saver = Saver(opts)
# train
print('\n--- train ---')
max_it = 500000
for ep in range(ep0, opts.n_ep):
for it, (images_a, images_b) in enumerate(train_loader):
if images_a.size(0) != opts.batch_size or images_b.size(
0) != opts.batch_size:
continue
# input data
images_a = images_a.cuda(opts.gpu).detach()
images_b = images_b.cuda(opts.gpu).detach()
# update model
if (it + 1) % opts.d_iter != 0 and it < len(train_loader) - 2:
model.update_D_content(images_a, images_b)
continue
else:
model.update_D(images_a, images_b)
model.update_EG()
# save to display file
if not opts.no_display_img and not opts.multi_modal:
saver.write_display(total_it, model)
print('total_it: %d (ep %d, it %d), lr %08f' %
(total_it, ep, it, model.gen_opt.param_groups[0]['lr']))
total_it += 1
if total_it >= max_it:
# saver.write_img(-1, model)
saver.write_model(-1, model)
break
# decay learning rate
if opts.n_ep_decay > -1:
model.update_lr()
# save result image
if not opts.multi_modal:
saver.write_img(ep, model)
# Save network weights
saver.write_model(ep, total_it, model)
return
def main():
args = parse_args()
speech_windowed_data, peak_distance, peak_indicator, indices, actual_gci_locations = create_dataset(
args.speechfolder, args.peaksfolder, args.window, args.stride, 10)
saver = Saver(args.model_dir)
model = SELUWeightNet
model, _, params_dict = saver.load_checkpoint(
model, file_name=args.model_name)
model.eval()
input = to_variable(
th.from_numpy(
np.expand_dims(speech_windowed_data, 1).astype(np.float32)),
args.use_cuda, True)
with warnings.catch_warnings():
if args.use_cuda:
model = model.cuda()
warnings.simplefilter('ignore')
prediction = model(input)
predicted_peak_indicator = F.sigmoid(prediction[:, 1]).data.numpy()
predicted_peak_distance = (prediction[:, 0]).data.numpy().astype(np.int32)
predicted_peak_indicator_indices = predicted_peak_indicator > args.threshold
predicted_peak_indicator = predicted_peak_indicator[
predicted_peak_indicator_indices].ravel()
predicted_peak_distance = predicted_peak_distance[
predicted_peak_indicator_indices].ravel()
indices = indices[predicted_peak_indicator_indices]
assert (len(indices) == len(predicted_peak_distance))
assert (len(predicted_peak_distance) == len(predicted_peak_indicator))
positive_distance_indices = predicted_peak_distance < args.window
positive_peak_distances = predicted_peak_distance[
positive_distance_indices]
postive_predicted_peak_indicator = predicted_peak_indicator[
positive_distance_indices]
print('Neg Peaks: {} Pos Peaks: {}'.format(
len(predicted_peak_distance) - len(positive_peak_distances),
len(positive_peak_distances)))
gci_locations = [
indices[i, d] for i, d in enumerate(positive_peak_distances)
]
locations_true = np.nonzero(actual_gci_locations)[0]
xaxes = np.zeros(len(actual_gci_locations))
xaxes[locations_true] = 1
if __debug__:
ground_truth = np.row_stack((np.arange(len(actual_gci_locations)),
xaxes))
predicted_truth = np.row_stack((gci_locations,
postive_predicted_peak_indicator))
os.makedirs(args.prediction_dir, exist_ok=True)
np.save(
os.path.join(args.prediction_dir, 'ground_truth'), ground_truth)
np.save(
os.path.join(args.prediction_dir, 'predicted'), predicted_truth)
plt.scatter(
gci_locations,
postive_predicted_peak_indicator,
color='b',
label='Predicted GCI')
plt.plot(
np.arange(len(actual_gci_locations)),
xaxes,
color='r',
label='Actual GCI')
plt.legend()
plt.show()
def __init__(self):
self.saver = Saver()
def main():
# parse options
parser = TrainOptions()
opts = parser.parse()
# daita loader
print('\n--- load dataset ---')
dataset = dataset_unpair(opts)
train_loader = torch.utils.data.DataLoader(dataset,
batch_size=opts.batch_size,
shuffle=True,
num_workers=opts.nThreads)
# model
print('\n--- load model ---')
model = UID(opts)
model.setgpu(opts.gpu)
if opts.resume is None:
model.initialize()
ep0 = -1
total_it = 0
else:
ep0, total_it = model.resume(opts.resume)
model.set_scheduler(opts, last_ep=ep0)
ep0 += 1
print('start the training at epoch %d' % (ep0))
# saver for display and output
saver = Saver(opts)
# train
print('\n--- train ---')
max_it = 500000
for ep in range(ep0, opts.n_ep):
for it, (images_a, images_b) in enumerate(train_loader):
if images_a.size(0) != opts.batch_size or images_b.size(
0) != opts.batch_size:
continue
images_a = images_a.cuda(opts.gpu).detach()
images_b = images_b.cuda(opts.gpu).detach()
# update model
model.update_D(images_a, images_b)
if (it + 1) % 2 != 0 and it != len(train_loader) - 1:
continue
model.update_EG()
# save to display file
if (it + 1) % 48 == 0:
print('total_it: %d (ep %d, it %d), lr %08f' %
(total_it + 1, ep, it + 1,
model.gen_opt.param_groups[0]['lr']))
print(
'Dis_I_loss: %04f, Dis_B_loss %04f, GAN_loss_I %04f, GAN_loss_B %04f'
% (model.disA_loss, model.disB_loss, model.gan_loss_i,
model.gan_loss_b))
print('B_percp_loss %04f, Recon_II_loss %04f' %
(model.B_percp_loss, model.l1_recon_II_loss))
if (it + 1) % 200 == 0:
saver.write_img(ep * len(train_loader) + (it + 1), model)
total_it += 1
if total_it >= max_it:
saver.write_img(-1, model)
saver.write_model(-1, model)
break
# decay learning rate
if opts.n_ep_decay > -1:
model.update_lr()
# Save network weights
saver.write_model(ep, total_it + 1, model)
return
def main():
# parse options
parser = TestOptions()
opts = parser.parse()
orig_dir = opts.orig_dir
blur_dir = opts.dataroot
saver = Saver(opts)
# data loader
print('\n--- load dataset ---')
dataset_domain = 'A' if opts.a2b else 'B'
# dataset = dataset_single(opts, 'A', opts.input_dim_a)
# else:
# dataset = dataset_single(opts, 'B', opts.input_dim_b)
# loader = torch.utils.data.DataLoader(dataset, batch_size=1, num_workers=opts.nThreads)
loader = CreateDataLoader(opts)
# model
print('\n--- load model ---')
model = UID(opts)
model.setgpu(opts.gpu) ## comment for cpu mode
model.resume(opts.resume, train=False)
model.eval()
# test
print('\n--- testing ---')
for idx1, data in enumerate(loader):
# img1, img_name_list = data[dataset_domain], data[dataset_domain+'_paths']
# img1 = img1.cuda(opts.gpu).detach()
images_b = data['B']
images_a = images_b # should in the same shape (This is only for the case `resize_or_crop="none"`)
img_name_list = data['B_paths'] # B is the fluorescence image
center_crop_shape = data[
'B_size_WH'][::-1] # B is the fluorescence image
if len(img_name_list) > 1:
print("Warning, there are more than 1 sample in the test batch.")
images_a = images_a.cuda(opts.gpu).detach() ## comment for cpu mode
images_b = images_b.cuda(opts.gpu).detach() ## comment for cpu mode
images_a = torch.cat(
[images_a] * 2,
dim=0) # because half of the batch is used as real_A_random
images_b = torch.cat(
[images_b] * 2,
dim=0) # because half of the batch is used as real_B_random
print('{}/{}'.format(idx1, len(loader)))
with torch.no_grad():
model.inference(images_a, images_b)
# img = model.test_forward(img1, a2b=opts.a2b)
img_name = img_name_list[0].split('/')[-1]
saver.write_img(idx1,
model,
img_name=img_name,
inference_mode=True,
mask_path='../input/testB_mask/' + img_name) # True
# saver.save_img(img=model.fake_I_encoded[[np.s_[:]]*2 + return_center_crop_slices(input_shapes=images_b.shape[-2:],
# output_shapes=center_crop_shape,
# input_scale=1.0,
# output_scale=opts.fineSize*1.0/opts.loadSize)],
# img_name=img_name,
# subfolder_name="fake_A") #'gen_%05d.png' % (idx1),
return
class Session(object):
L, R, X = 0, 1, 2
COR, INCOR, EARLY, BOTH, NULL, KILLED, SKIPPED = 0, 1, 2, 3, 4, 5, 6
PHASE_INTRO, PHASE_STIM, PHASE_DELAY, PHASE_LICK, PHASE_REWARD, PHASE_ITI, PHASE_END = [0, 1, 2, 3, 4, 5, 6]
DEFAULT_PARAMS = dict(
n_trials=[100], # if a list, corresponds to list of lams
lam=[
0.5
], # poisson lambda for one side, as fraction of n_total_stims. ex 0.9 means that on any given trial, the expected number of stims on one side will be 0.9*n_total_stims and the expected number of stims on the other side will be 0.1*n_total_stims. If a list, will be sequential trials with those params, numbers of trials specified by n_trials parameter. If any item in this list is a list, it means multiple shuffled trials equally distributed across those lams.
# n_total_stims = 10.,
rate_sum=5.0,
max_rate_frac=2.0,
# max_n_stims = 20, #on one side
stim_duration=0.050,
stim_duration_override=False,
stim_phase_duration=[5.0, 1.0], # mean, std
enforce_stim_phase_duration=True,
stim_phase_intro_duration=0.200,
stim_phase_end_duration=0.050,
min_isi=0.100,
reward_duration=[0.010, 0.010],
penalty_iti_frac=1.0, # fraction of ITI to add to normal ITI when trial was incorrect
distribution_mode="poisson",
phase_durations={
PHASE_INTRO: 1.0,
PHASE_STIM: None,
PHASE_DELAY: 1.0,
PHASE_LICK: 4.0,
PHASE_REWARD: 4.0,
PHASE_ITI: 3.0,
PHASE_END: 0.0,
}, # PHASE_END must always have 0.0 duration
iter_resolution=0.030,
hold_rule=True,
puffs_on=True,
rewards_on=True,
hints_on=False, # False, or n first trials to give hints, or True for all trials
hint_interval=0.500,
bias_correction=True,
bias_correction_window=6,
max_bias_correction=0.1,
lick_rule_phase=True, # must not lick before the lick phase
lick_rule_side=True, # must not lick incorrect side during the lick phase
lick_rule_any=True, # must lick some port in lick phase to get reward
multiple_decisions=False, # can make multiple attempts at correct side
use_trials=True, # if false, trials are ignored. used for training
trial_vid_freq=2, # movie for every n trials
extra_bigdiff_trials=False,
condition=-1,
)
def __init__(self, subj, session_params, cam_params=default_cam_params):
self.params = self.DEFAULT_PARAMS
self.cam_params = cam_params
self.subj = subj
self.name = time.strftime("%Y%m%d_%H%M%S")
self.saver = Saver(self.subj, self.name)
# kwargs
self.params.update(session_params)
# checks on kwargs
assert self.params["min_isi"] > self.params["stim_duration"] # otherwise same-side stims can overlap
if self.params["lick_rule_side"]:
assert self.params[
"lick_rule_any"
] # if must lick correct side in lick phase, then must lick at all in lick phase
# extensions of kwargs
if not isinstance(self.params["lam"], list):
self.params["lam"] = [self.params["lam"]]
if not isinstance(self.params["n_trials"], list):
self.params["n_trials"] = [self.params["n_trials"]]
assert len(self.params["lam"]) == len(self.params["n_trials"])
self.params["subj_name"] = self.subj.name
# self.params['phase_durations'][self.PHASE_STIM] = self.params['stim_phase_intro_duration'] + self.params['stim_phase_duration']
self.cam_params.update(dict(save_name=pjoin(self.subj.subj_dir, self.name)))
if self.params["hints_on"] is True:
self.params["hints_on"] = 1e6
# add them in
self.__dict__.update(self.params)
# hardware
syncobj = multiprocessing.Value("d", 0)
threading.Thread(target=update_sync, args=(syncobj,)).start()
self.cam = PSEye(clock_sync_obj=syncobj, **self.cam_params)
self.cam.start()
self.lr = AnalogReader(
saver=self.saver, lick_thresh=5.0, ports=["ai0", "ai1", "ai5", "ai6"], runtime_ports=[0, 1]
)
sd = self.stim_duration
if self.stim_duration_override:
sd = self.stim_duration_override
self.stimulator = Valve(saver=self.saver, ports=["port0/line0", "port0/line1"], name="stimulator", duration=sd)
self.spout = Valve(
saver=self.saver, ports=["port0/line2", "port0/line3"], name="spout", duration=self.reward_duration
)
self.light = Light(saver=self.saver, port="ao0")
self.speaker = Speaker(saver=self.saver)
# trials init
self.trials = self.generate_trials()
# save session info
self.saver.save_session(self)
# runtime variables
self.session_on = 0
self.session_complete = False
self.session_kill = False
self.trial_on = 0
self.trial_idx = -1
self.valid_trial_idx = 0
self.saving_trial_idx = 0
self.session_runtime = -1
self.trial_runtime = -1
self.trial_outcomes = []
self.trial_corrects = [] # for use in use_trials==False
self.rewards_given = 0
self.paused = 0
self.holding = False
self.percentages = [0.0, 0.0] # L/R
self.side_ns = [0, 0] # L/R
self.bias_correction_percentages = [0.0, 0.0] # L/R
self.perc_valid = 0.0
self.iter_write_begin = now()
def generate_trials(self):
timess = []
lamss = []
durs = []
for lam, n_trials in zip(self.lam, self.n_trials):
for _ in xrange(n_trials):
if isinstance(lam, list):
lami = np.random.choice(lam)
else:
lami = lam
lams = [lami, 1.0 - lami]
lamss.append(lams)
dur = np.random.normal(*self.stim_phase_duration)
durs.append(dur)
tt = self.generate_stim_times(
lams=lams, dur=dur, intro_dur=self.stim_phase_intro_duration, min_isi=self.min_isi
)
if self.extra_bigdiff_trials and 1.0 in lams and np.random.random() < 0.7:
enforce_n = np.random.choice([11, 12, 13, 14, 15])
while max([len(i) for i in tt]) < enforce_n:
tt = self.generate_stim_times(
lams=lams, dur=dur, intro_dur=self.stim_phase_intro_duration, min_isi=self.min_isi
)
timess.append(tt)
trials = np.zeros(
(len(timess)), dtype=[("times", vlen_array_dtype), ("correct", int), ("lams", float, 2), ("dur", float)]
)
lr_idxer = [0, 1]
for tidx, times, lams, d in zip(range(len(timess)), timess, lamss, durs):
np.random.shuffle(lr_idxer)
times = [times[lr_idxer[0]], times[lr_idxer[1]]] # left or right equally likely to correspond to lam[0]
lams = [lams[lr_idxer[0]], lams[lr_idxer[1]]] # maintain alignment with lams
lenL, lenR = [len(times[i]) for i in [self.L, self.R]]
if lenL > lenR:
correct = self.L
elif lenR > lenL:
correct = self.R
elif lenR == lenL:
raise Exception("Indeterminate trial included. This should not be possible!")
trials[tidx] = (times, correct, lams, d)
return np.array(trials)
def generate_stim_times(self, lams, dur, intro_dur, min_isi, time_resolution=0.001):
n_stim = [0, 0]
while n_stim[0] == n_stim[1] or np.max(n_stim) / dur > self.rate_sum * self.max_rate_frac:
if self.distribution_mode == "poisson":
n_stim = np.random.poisson(self.rate_sum * dur * np.asarray(lams))
elif self.distribution_mode == "abs":
n_stim = np.round(self.n_total_stims * np.asarray(lams))
t = np.arange(intro_dur, intro_dur + dur + time_resolution, time_resolution)
def make(sz):
times = np.append(0, np.sort(np.random.choice(t, size=sz, replace=False)))
while len(times) > 1 and np.diff(times).min() < min_isi:
times = np.append(0, np.sort(np.random.choice(t, size=sz, replace=False)))
return times
stim_times = map(make, n_stim)
return np.array(stim_times)
def get_cum_performance(self):
cum = np.asarray(self.trial_outcomes)
markers = cum.copy()
if self.lick_rule_phase:
ignore = [self.SKIPPED, self.EARLY, self.NULL, self.KILLED]
else:
ignore = [self.SKIPPED, self.NULL, self.KILLED]
valid = np.array([c not in ignore for c in cum]).astype(bool)
cum = cum == self.COR
cum = [
np.mean([c for c, v in zip(cum[:i], valid[:i]) if v]) if np.any(valid[:i]) else 0.0
for i in xrange(1, len(cum) + 1)
] # cumulative
return cum, markers, np.asarray(self.trial_corrects)
def stimulate(self, trial):
sides = [self.L, self.R]
np.random.shuffle(sides)
for side in sides:
tr = trial["times"][side]
si = self.stim_idx[side]
if si >= len(tr):
return
dt_phase = now() - self.phase_start
if dt_phase >= tr[si]:
self.stimulator.go(side)
self.stim_idx[side] += 1
if np.all(np.asarray(self.stim_idx) == np.asarray(map(len, trial["times"]))):
self.stim_complete = True
def to_phase(self, ph):
self.phase_times[self.current_phase][1] = now()
self.phase_times[ph][0] = now()
self.current_phase = ph
self.phase_start = now()
self.hinted = False
if ph == self.PHASE_END:
# sanity check. should have been rewarded only if solely licked on correct side
if (
self.lick_rule_side
and self.lick_rule_phase
and (not self.licked_early)
and (not self.multiple_decisions)
and self.use_trials
):
assert bool(self.rewarded) == (
any(self.lick_phase_licks[self.trial["correct"]])
and (not any(self.lick_phase_licks[-self.trial["correct"] + 1]))
)
if not self.rewarded:
if self.use_trials:
self.trial_corrects.append(self.trial["correct"])
else:
self.trial_corrects.append(self.X)
# determine trial outcome
if not self.use_trials:
if any(self.lick_phase_licks):
outcome = self.COR
else:
outcome = self.INCOR
elif self.use_trials:
if self.rewarded and self.lick_rule_side and not self.multiple_decisions:
outcome = self.COR
elif self.rewarded and ((not self.lick_rule_side) or self.multiple_decisions):
lpl_min = np.array([min(i) if len(i) else -1 for i in self.lickph_andon_licks])
if np.all(lpl_min == -1):
outcome = self.NULL
else:
lpl_min[lpl_min == -1] = now()
if np.argmin(lpl_min) == self.trial["correct"]:
outcome = self.COR
else:
outcome = self.INCOR
elif self.trial_kill:
outcome = self.KILLED
elif self.licked_early:
outcome = self.EARLY
# this BOTH logic no longer works bc i include reward phase licks in lickphaselicks. both will never show up, though it still can *rarely* be a cause for trial failure
# elif (any(self.lick_phase_licks[self.L]) and any(self.lick_phase_licks[self.R])):
# outcome = self.BOTH
elif any(self.lick_phase_licks[-self.trial["correct"] + 1]):
outcome = self.INCOR
elif not any(self.lick_phase_licks):
outcome = self.NULL
self.trial_outcomes.append(outcome)
def update_licked(self):
l = self.lr.licked()
tst = now()
for idx, li in enumerate(l):
self.licks[idx] += [
tst
] * li # represent not the number of licks but the number of times the LR class queried daq and found a positive licking signal
if self.current_phase in [self.PHASE_LICK, self.PHASE_REWARD]:
self.lickph_andon_licks[idx] += [tst] * li
if self.current_phase == self.PHASE_LICK:
self.lick_phase_licks[idx] += [tst] * li
if self.hold_rule:
if (not self.holding) and np.any(self.lr.holding):
self.holding = True
self.paused += 1
elif self.holding and not np.any(self.lr.holding):
self.paused = max(self.paused - 1, 0)
self.holding = False
if self.holding:
self.speaker.pop()
def run_phase(self, ph):
ph_dur = self.phase_durations[ph] # intended phase duration
if ph == self.PHASE_STIM:
ph_dur = (
self.stim_phase_intro_duration + self.trial["dur"] + self.stim_phase_end_duration
) # before dur was introduced: np.max(np.concatenate(self.trial['times']))+self.stim_phase_end_duration
dt_phase = now() - self.phase_start
self.session_runtime = now() - self.session_on
self.trial_runtime = now() - self.trial_on
self.update_licked()
# special cases
if ph == self.PHASE_ITI and not self.rewarded:
ph_dur *= 1 + self.penalty_iti_frac
if now() - self.iter_write_begin >= self.iter_resolution:
iter_info = dict(
trial=self.trial_idx,
phase=ph,
licks=np.array([len(i) for i in self.licks]),
licked_early=self.licked_early,
dt_phase=dt_phase,
paused=self.paused,
)
self.saver.write("iterations", iter_info)
self.iter_write_begin = now()
if self.paused and self.current_phase in [self.PHASE_INTRO, self.PHASE_STIM, self.PHASE_DELAY, self.PHASE_LICK]:
self.trial_kill = True
return
if self.trial_kill and not self.current_phase == self.PHASE_ITI:
self.to_phase(self.PHASE_ITI)
return
# Intro
if ph == self.PHASE_INTRO:
self.light.off()
if not self.intro_signaled:
self.speaker.intro()
self.intro_signaled = True
# comment out to give intro time with licks allowed? doesnt work, cause lick variable accumulates over trial
if any(self.licks) and self.lick_rule_phase:
self.licked_early = min(self.licks[0] + self.licks[1])
self.to_phase(self.PHASE_ITI)
return
if dt_phase >= ph_dur:
self.to_phase(self.PHASE_STIM)
return
# Stim
elif ph == self.PHASE_STIM:
if any(self.licks) and self.lick_rule_phase:
self.licked_early = min(self.licks[0] + self.licks[1])
self.to_phase(self.PHASE_ITI)
return
if dt_phase >= ph_dur:
self.to_phase(self.PHASE_DELAY)
return
if self.puffs_on:
self.stimulate(self.trial)
if (not self.enforce_stim_phase_duration) and self.stim_complete:
self.to_phase(self.PHASE_DELAY)
return
# Delay
elif ph == self.PHASE_DELAY:
if any(self.licks) and self.lick_rule_phase:
self.licked_early = min(self.licks[0] + self.licks[1])
self.to_phase(self.PHASE_ITI)
return
if dt_phase >= ph_dur:
self.to_phase(self.PHASE_LICK)
return
# Lick
elif ph == self.PHASE_LICK:
self.light.on()
if (
self.hints_on
and self.hints_on > self.trial_idx
and (not self.hinted or (now() - self.hinted) > self.hint_interval)
and self.use_trials
):
self.stimulator.go(self.trial["correct"])
self.hinted = now()
if not self.laser_signaled:
self.speaker.laser()
self.laser_signaled = True
if not self.lick_rule_any:
self.to_phase(self.PHASE_REWARD)
return
if any(self.lick_phase_licks) and not self.multiple_decisions:
self.to_phase(self.PHASE_REWARD)
return
elif (
any(self.lick_phase_licks)
and self.multiple_decisions
and any(self.lick_phase_licks[self.trial["correct"]])
):
self.to_phase(self.PHASE_REWARD)
return
# if time is up, to reward phase
if dt_phase >= ph_dur:
self.to_phase(self.PHASE_REWARD)
return
# Reward
elif ph == self.PHASE_REWARD:
self.light.on() # probably redundant
if (
self.lick_rule_side
and any(self.lick_phase_licks[-self.trial["correct"] + 1])
and not self.rewarded
and not self.multiple_decisions
):
self.speaker.wrong()
self.to_phase(self.PHASE_ITI)
return
# sanity check. cannot reach here if any incorrect licks, ensure that:
if self.lick_rule_side and not self.multiple_decisions:
assert not any(self.lick_phase_licks[-self.trial["correct"] + 1])
# if no licks at all, go straight to ITI
if self.lick_rule_any and not any(self.lick_phase_licks):
self.to_phase(self.PHASE_ITI)
return
# if allowed multiple choices but only licked wrong side
if self.multiple_decisions and not any(self.lick_phase_licks[self.trial["correct"]]):
self.to_phase(self.PHASE_ITI)
return
# sanity check. can only reach here if licked correct side only
if self.lick_rule_any and self.lick_rule_side:
assert any(self.lick_phase_licks[self.trial["correct"]])
# from this point on, it is assumed that rewarding should occur.
if self.use_trials:
rside = self.trial["correct"]
else:
rside = np.argmin([min(i) if len(i) else now() for i in self.lick_phase_licks])
if not self.corside_added:
self.trial_corrects.append(rside)
self.corside_added = True
if (
self.hints_on
and self.hints_on > self.trial_idx
and (not self.hinted or (now() - self.hinted) > self.hint_interval)
):
self.stimulator.go(rside)
self.hinted = now()
if self.rewards_on and not self.rewarded:
self.spout.go(side=rside)
self.rewarded = now()
self.rewards_given += 1
if dt_phase >= ph_dur:
self.to_phase(self.PHASE_ITI)
# ITI
elif ph == self.PHASE_ITI:
self.light.off()
if self.licked_early and self.lick_rule_phase and not self.error_signaled:
self.speaker.error()
self.error_signaled = True
if dt_phase >= ph_dur:
self.to_phase(self.PHASE_END)
return
def determine_skip(self):
to = np.asarray(self.trial_outcomes)
if len(to) == 0:
return False
corincor_idxs = np.where(np.logical_or(to == self.COR, to == self.INCOR))[0]
all_val = np.sum([i in [self.COR, self.INCOR, self.EARLY, self.NULL, self.KILLED] for i in to])
if all_val != 0:
self.perc_valid = float(len(corincor_idxs)) / all_val
corincor_trials = self.trials[corincor_idxs]
trcor = corincor_trials["correct"]
subcor = to[corincor_idxs] == self.COR
if np.sum(trcor == self.L) > 0:
perc_l = np.mean(subcor[trcor == self.L])
self.percentages[self.L] = perc_l
self.side_ns[self.L] = np.sum(trcor == self.L)
if np.sum(trcor == self.R) > 0:
perc_r = np.mean(subcor[trcor == self.R])
self.percentages[self.R] = perc_r
self.side_ns[self.R] = np.sum(trcor == self.R)
if (
np.sum(trcor == self.L) < self.bias_correction_window
or np.sum(trcor == self.R) < self.bias_correction_window
):
return False
perc_l = np.mean(subcor[trcor == self.L][-self.bias_correction_window :])
perc_r = np.mean(subcor[trcor == self.R][-self.bias_correction_window :])
self.bias_correction_percentages = [perc_l, perc_r]
if perc_l == perc_r:
return False
this_cor = self.trials[self.trial_idx]["correct"]
if self.bias_correction_percentages[this_cor] < self.bias_correction_percentages[-1 + this_cor]:
return False
if min(self.bias_correction_percentages) == 0:
pthresh = self.max_bias_correction
else:
pthresh = float(min(self.bias_correction_percentages)) / max(self.bias_correction_percentages)
return np.random.random() > pthresh
def next_trial(self):
# init the trial
self.skip_trial = False
self.trial_idx += 1
if self.trial_idx > 1 and self.trial_outcomes[-1] in [self.COR, self.INCOR]:
self.valid_trial_idx += 1
if self.trial_idx >= len(self.trials):
self.session_complete = True
return
self.skip_trial = self.determine_skip()
self.trial_on = now()
self.trial = self.trials[self.trial_idx]
self.current_phase = self.PHASE_INTRO
self.stim_idx = [0, 0] # index of next stim, for [L,R]
self.trial_start = now()
self.phase_start = self.trial_start
self.phase_times = [[-1, -1] for _ in self.phase_durations]
_ = self.lr.licked() # to clear any residual signal
self.licks = [[], []]
self.lick_phase_licks = [[], []]
self.lickph_andon_licks = [[], []]
self.licked_early = False
self.rewarded = False
self.error_signaled = False
self.laser_signaled = False
self.intro_signaled = False
self.stim_complete = False
self.trial_kill = False
self.hinted = False
self.corside_added = False
if self.skip_trial:
self.cam.SAVING.value = 0
else:
if self.saving_trial_idx % self.trial_vid_freq == 0:
self.cam.SAVING.value = 1
else:
self.cam.SAVING.value = 0
self.saving_trial_idx += 1
# logging.info('Starting trial %i.' %self.trial_idx)
# run the trial loop
if self.skip_trial:
self.trial_outcomes.append(self.SKIPPED)
self.trial_corrects.append(self.trial["correct"])
else:
while self.current_phase != self.PHASE_END:
self.run_phase(self.current_phase)
# save trial info
trial_info = dict(
idx=self.trial_idx,
ns=[len(tt) for tt in self.trial["times"]],
start_time=self.trial_start,
licksL=self.licks[self.L],
licksR=self.licks[self.R],
licked_early=self.licked_early,
phase_times=self.phase_times,
rewarded=self.rewarded,
outcome=self.trial_outcomes[-1],
hints=(self.hints_on and self.hints_on > self.trial_idx),
end_time=now(),
condition=self.condition,
)
self.saver.write("trials", trial_info)
def run(self):
self.session_on = now()
self.lr.begin_saving()
while True:
self.next_trial()
if self.session_kill:
logging.info("Session killed manually")
self.paused = False
break
if self.session_complete:
logging.info("Session complete")
break
self.session_on = False
self.end()
def end(self):
to_end = [self.lr, self.stimulator, self.spout, self.light, self.cam, self.saver]
for te in to_end:
te.end()
time.sleep(0.050)
def get_code(self):
py_files = [
pjoin(d, f) for d, _, fs in os.walk(os.getcwd()) for f in fs if f.endswith(".py") and not f.startswith("__")
]
code = {}
for pf in py_files:
with open(pf, "r") as f:
code[pf] = f.read()
return json.dumps(code)
def __init__(self, master=None):
Frame.__init__(self, master)
self.master.title("Diary Prompter")
self.init_elements()
self.saver = Saver()
sweep.attox = attox
sweep.attoy = attoy
sweep.attoz = attoz
sweep.daq = daq
i = it.Foo()
print(daq.__class__)
print(daq.__class__.__name__)
print(i.__class__)
print(i.__class__.__name__)
print('\n[dict2, objlist2] = attox.todict([])')
[dict2, objlist2] = attox.todict([])
print('dict2: \n' + str(Saver.tocommentjson(dict2)))
print('objlist2: ' + str(objlist2))
print('\n[dict3, objlist3] = attoy.todict([])')
[dict3, objlist3] = attoy.todict([])
print('dict3: \n' + str(Saver.tocommentjson(dict3)))
print('objlist3: ' + str(objlist3))
print('\n[dict4, objlist4] = attoz.todict([])')
[dict4, objlist4] = attoz.todict([])
print('dict4: \n' + str(Saver.tocommentjson(dict4)))
print('objlist4: ' + str(objlist4))
print('\n[dict1, objlist1] = daq.todict([])')
[dict1, objlist1] = daq.todict([])
print('dict1: \n' + str(Saver.tocommentjson(dict1)))
def validate(val_dirs: ValidationDirs, images_iterator: ImagesIterator,
flags: OutputFlags):
"""
Saves in val_dirs.log_dir/val/dataset_name/measures.csv:
- `img_name,bpp,psnr,ms-ssim forall img_name`
"""
print(_VALIDATION_INFO_STR)
validated_checkpoints = val_dirs.get_validated_checkpoints(
) # :: [10000, 18000, ..., 256000], ie, [int]
all_ckpts = Saver.all_ckpts_with_iterations(val_dirs.ckpt_dir)
if len(all_ckpts) == 0:
print('No checkpoints found in {}'.format(val_dirs.ckpt_dir))
return
# if ckpt_step is -1, then all_ckpt[:-1:flags.ckpt_step] === [] because of how strides work
ckpt_to_check = all_ckpts[:-1:flags.ckpt_step] + [
all_ckpts[-1]
] # every ckpt_step-th checkpoint plus the last one
if flags.ckpt_step == -1:
assert len(ckpt_to_check) == 1
print('Validating {}/{} checkpoints (--ckpt_step {})...'.format(
len(ckpt_to_check), len(all_ckpts), flags.ckpt_step))
missing_checkpoints = [(ckpt_itr, ckpt_path)
for ckpt_itr, ckpt_path in ckpt_to_check
if ckpt_itr not in validated_checkpoints]
if len(missing_checkpoints) == 0:
print('All checkpoints validated, stopping...')
return
# ---
# create networks
autoencoder_config_path, probclass_config_path = logdir_helpers.config_paths_from_log_dir(
val_dirs.log_dir,
base_dirs=[constants.CONFIG_BASE_AE, constants.CONFIG_BASE_PC])
ae_config, ae_config_rel_path = config_parser.parse(
autoencoder_config_path)
pc_config, pc_config_rel_path = config_parser.parse(probclass_config_path)
ae_cls = autoencoder.get_network_cls(ae_config)
pc_cls = probclass.get_network_cls(pc_config)
# Instantiate autoencoder and probability classifier
ae = ae_cls(ae_config)
pc = pc_cls(pc_config, num_centers=ae_config.num_centers)
x_val_ph = tf.placeholder(tf.uint8, (3, None, None), name='x_val_ph')
x_val_uint8 = tf.expand_dims(x_val_ph, 0, name='batch')
x_val = tf.to_float(x_val_uint8, name='x_val')
enc_out_val = ae.encode(x_val, is_training=False)
x_out_val = ae.decode(enc_out_val.qhard, is_training=False)
bc_val = pc.bitcost(enc_out_val.qbar,
enc_out_val.symbols,
is_training=False,
pad_value=pc.auto_pad_value(ae))
bpp_val = bits.bitcost_to_bpp(bc_val, x_val)
x_out_val_uint8 = tf.cast(x_out_val, tf.uint8, name='x_out_val_uint8')
# Using numpy implementation due to dynamic shapes
msssim_val = ms_ssim_np.tf_msssim_np(x_val_uint8,
x_out_val_uint8,
data_format='NCHW')
psnr_val = psnr_np(x_val_uint8, x_out_val_uint8)
restorer = Saver(val_dirs.ckpt_dir,
var_list=Saver.get_var_list_of_ckpt_dir(
val_dirs.ckpt_dir))
# create fetch_dict
fetch_dict = {
'bpp': bpp_val,
'ms-ssim': msssim_val,
'psnr': psnr_val,
}
if flags.real_bpp:
fetch_dict['sym'] = enc_out_val.symbols # NCHW
if flags.save_ours:
fetch_dict['img_out'] = x_out_val_uint8
# ---
fw = tf.summary.FileWriter(val_dirs.out_dir, graph=tf.get_default_graph())
def full_summary_tag(summary_name):
return '/'.join(['val', images_iterator.dataset_name, summary_name])
# Distance
try:
codec_distance_ms_ssim = CodecDistance(images_iterator.dataset_name,
codec='bpg',
metric='ms-ssim')
codec_distance_psnr = CodecDistance(images_iterator.dataset_name,
codec='bpg',
metric='psnr')
except CodecDistanceReadException as e: # no codec distance values stored for the current setup
print('*** Distance to BPG not available for {}:\n{}'.format(
images_iterator.dataset_name, str(e)))
codec_distance_ms_ssim = None
codec_distance_psnr = None
# Note that for each checkpoint, the structure of the network will be the same. Thus the pad depending image
# loading can be cached.
# create session
with tf_helpers.create_session() as sess:
if flags.real_bpp:
pred = probclass.PredictionNetwork(pc, pc_config,
ae.get_centers_variable(), sess)
checker = probclass.ProbclassNetworkTesting(pc, ae, sess)
bpp_fetcher = bpp_helpers.BppFetcher(pred, checker)
fetcher = sess.make_callable(fetch_dict, feed_list=[x_val_ph])
last_ckpt_itr = missing_checkpoints[-1][0]
for ckpt_itr, ckpt_path in missing_checkpoints:
if not ckpt_still_exists(ckpt_path):
# May happen if job is still training
print('Checkpoint disappeared: {}'.format(ckpt_path))
continue
print(_CKPT_ITR_INFO_STR.format(ckpt_itr))
restorer.restore_ckpt(sess, ckpt_path)
values_aggregator = ValuesAggregator('bpp', 'ms-ssim', 'psnr')
# truncates the previous measures.csv file! This way, only the last valid checkpoint is saved.
measures_writer = MeasuresWriter(val_dirs.out_dir)
# ----------------------------------------
# iterate over images
# images are padded to work with current auto encoder
for img_i, (img_name, img_content) in enumerate(
images_iterator.iter_imgs(
pad=ae.get_subsampling_factor())):
otp = fetcher(img_content)
measures_writer.append(img_name, otp)
if flags.real_bpp:
# Calculate
bpp_real, bpp_theory = bpp_fetcher.get_bpp(
otp['sym'],
bpp_helpers.num_pixels_in_image(img_content))
# Logging
bpp_loss = otp['bpp']
diff_percent_tr = (bpp_theory / bpp_real) * 100
diff_percent_lt = (bpp_loss / bpp_theory) * 100
print('BPP: Real {:.5f}\n'
' Theoretical: {:.5f} [{:5.1f}% of real]\n'
' Loss: {:.5f} [{:5.1f}% of real]'.format(
bpp_real, bpp_theory, diff_percent_tr, bpp_loss,
diff_percent_lt))
assert abs(
bpp_theory - bpp_loss
) < 1e-3, 'Expected bpp_theory to match loss! Got {} and {}'.format(
bpp_theory, bpp_loss)
if flags.save_ours and ckpt_itr == last_ckpt_itr:
save_img(img_name, otp['img_out'], val_dirs)
values_aggregator.update(otp)
print('{: 10d} {img_name} | Mean: {avgs}'.format(
img_i,
img_name=img_name,
avgs=values_aggregator.averages_str()),
end=('\r' if not flags.real_bpp else '\n'),
flush=True)
measures_writer.close()
print() # add newline
avgs = values_aggregator.averages()
avg_bpp, avg_ms_ssim, avg_psnr = avgs['bpp'], avgs[
'ms-ssim'], avgs['psnr']
tf_helpers.log_values(
fw, [(full_summary_tag('avg_bpp'), avg_bpp),
(full_summary_tag('avg_ms_ssim'), avg_ms_ssim),
(full_summary_tag('avg_psnr'), avg_psnr)],
iteration=ckpt_itr)
if codec_distance_ms_ssim and codec_distance_psnr:
try:
d_ms_ssim = codec_distance_ms_ssim.distance(
avg_bpp, avg_ms_ssim)
d_pnsr = codec_distance_psnr.distance(avg_bpp, avg_psnr)
print('Distance to BPG: {:.3f} ms-ssim // {:.3f} psnr'.
format(d_ms_ssim, d_pnsr))
tf_helpers.log_values(
fw,
[(full_summary_tag('distance_BPG_MS-SSIM'), d_ms_ssim),
(full_summary_tag('distance_BPG_PSNR'), d_pnsr)],
iteration=ckpt_itr)
except ValueError as e: # out of range errors from distance calls
print(e)
val_dirs.add_validated_checkpoint(ckpt_itr)
print('Validation completed {}'.format(val_dirs))
from bone import Bone
from game_over import GameOver
from get_key import ClearConsole, GetKey, Wait, WindowSize
from laser import Laser
from map import Map
from paco import Paco
from saver import Saver
from start import Start
if __name__ == '__main__':
map = Map(133, 33, '#')
start = Start(133, 33)
finish = GameOver(133, 33)
paconator = Paco(1, 15, 133, 33)
getKey = GetKey()
save = Saver()
pigeons = []
lasers = []
bones = []
play_game = False
game_over = False
ammo = 20
points = 0
x = save.load()
record = int(x)
WindowSize()
# ====== START SCREEN ======
while True:
key = getKey()
for y in range(11):
def __init__(self, json_data, begin, end, directory):
Saver.__init__(self, json_data, directory, QzonePath.BLOG)
self._filename = "blogs_%05d-%05d.json" % (begin, end)
def train(self):
# saver
self.logger.info('Initialize saver ...')
train_saver = Saver(self.sess, tf.global_variables(),
self.cfg.model_dir)
merged = tf.summary.merge_all()
writer = tf.summary.FileWriter(self.cfg.log_dir, self.sess.graph)
# initialize weights
self.logger.info('Initialize all variables ...')
self.sess.run(
tf.variables_initializer(tf.global_variables(), name='init'))
self.load_weights(self.cfg.init_model)
self.logger.info('Start training ...')
start_itr = self.cur_epoch * self.itr_per_epoch + 1
end_itr = self.itr_per_epoch * self.cfg.end_epoch + 1
for itr in range(start_itr, end_itr):
self.tot_timer.tic()
self.cur_epoch = itr // self.itr_per_epoch
setproctitle.setproctitle('train epoch:' + str(self.cur_epoch))
cur_lr = get_lr(self.cur_epoch)
if not approx_equal(cur_lr, self.lr_eval):
print(self.lr_eval, cur_lr)
self.sess.run(tf.assign(self.lr, cur_lr))
# input data
self.read_timer.tic()
feed_dict = self.next_feed()
self.read_timer.toc()
# train one step
self.gpu_timer.tic()
_, self.lr_eval, *summary_res = self.sess.run(
[self.graph_ops[0], self.lr, *self.summary_dict.values()],
feed_dict=feed_dict)
self.gpu_timer.toc()
if (itr % 1 == 0):
result = self.sess.run(merged, feed_dict=feed_dict)
writer.add_summary(result, itr)
itr_summary = dict()
for i, k in enumerate(self.summary_dict.keys()):
itr_summary[k] = summary_res[i]
screen = [
'Epoch %d itr %d/%d:' %
(self.cur_epoch, itr, self.itr_per_epoch),
'lr: %g' % (self.lr_eval),
'speed: %.2f(%.2fs r%.2f)s/itr' %
(self.tot_timer.average_time, self.gpu_timer.average_time,
self.read_timer.average_time),
'%.2fh/epoch' %
(self.tot_timer.average_time / 3600. * self.itr_per_epoch),
' '.join(
map(lambda x: '%s: %.4f' % (x[0], x[1]),
itr_summary.items())),
]
if itr % self.cfg.display == 0:
self.logger.info(' '.join(screen))
if itr % self.itr_per_epoch == 0:
train_saver.save_model(self.cur_epoch)
self.tot_timer.toc()
rel_err = torch.abs(x_pred_true - x_output_true) / x_output_true
return torch.sum(abs_err) / torch.nonzero(binary_mask).size(0), torch.sum(rel_err) / torch.nonzero(binary_mask).size(0)
##############################################################################
##############################################################################
################################ MAIN ################################
##############################################################################
##############################################################################
# Saving settings
model_dir = os.path.join(opt.checkpoint_path, opt.name)
os.mkdir(model_dir) if not os.path.isdir(model_dir) else None
saver = Saver(model_dir, args=opt)
# Define model and optimiser
gpu = utils.check_gpu()
device = torch.device("cuda:{}".format(gpu) if torch.cuda.is_available() else "cpu")
model = SegNet().to(device)
optimizer = optim.Adam(model.parameters(), lr=1e-4)
# Recover weights, if required
if opt.recover:
ckpt_file = os.path.join(model_dir, opt.reco_type+'_weights.pth')
ckpt = torch.load(ckpt_file, map_location=device)
model.load_state_dict(ckpt['model_state_dict'])
epoch = ckpt['iter_nb'] + 1
print('Model recovered from {}.'.format(ckpt_file))
def saver_work(args, queues):
# with tf.device('/gpu:0'):
saver = Saver(args, queues)
saver.run()
