def main(args):
# Create model directory for saving trained models
if not os.path.exists(args.model_path):
os.makedirs(args.model_path)
test_dataset = a2d_dataset.A2DDataset(train_cfg, args.dataset_path)
data_loader = DataLoader(test_dataset,
batch_size=1,
shuffle=False,
num_workers=1)
# define load your model here
model_1 = net('efficientnet_b7')
model_1.cuda()
model_1.load_state_dict(
torch.load(os.path.join(args.model_path, 'efficientnetb7_F53.8.ckpt')))
model_2 = net('efficientnet_b7')
model_2.cuda()
model_2.load_state_dict(
torch.load(
os.path.join(args.model_path, 'efficientnetb7_val_53.7.ckpt')))
X = np.zeros((data_loader.__len__(), args.num_cls))
Y = np.zeros((data_loader.__len__(), args.num_cls))
print(data_loader.__len__())
model_1.eval()
model_2.eval()
with torch.no_grad():
for batch_idx, data in enumerate(data_loader):
# mini-batch
images = data[0].to(device)
labels = data[1].type(torch.FloatTensor).to(device)
# output = model(images).cpu().detach().numpy()
output_1 = model_1(images)
# output_1 = (torch.nn.functional.sigmoid(output_1)).cpu().detach().numpy()
output_2 = model_2(images)
# output_2 = (torch.nn.functional.sigmoid(output_2)).cpu().detach().numpy()
output = (output_1 + output_2) / 2
output = (
torch.nn.functional.sigmoid(output)).cpu().detach().numpy()
target = labels.cpu().detach().numpy()
output[output >= 0.4] = 1
output[output < 0.4] = 0
X[batch_idx, :] = output
Y[batch_idx, :] = target
P = Precision(X, Y)
R = Recall(X, Y)
F = F1(X, Y)
print('Precision: {:.1f} Recall: {:.1f} F1: {:.1f}'.format(
100 * P, 100 * R, 100 * F))
def cnfolCollectorProcess(startDateCursor='2018-03-01'):
zjNet = network.net()
if zjNet.loadCookie(__ZJ_MAINPAGE__) is None:
raise network.networkError("__ZJ_MAINPAGE__")
rf = record.recordfile(__RECORDFILE_ZJ__, __FIELDNAMES__)
cursor = startDateCursor
insertCusumError = 0
lockfile = open('./lockPage_zj','w+')
while cursor:
lockfile.seek(0)
lockfile.write(str(cursor))
lockfile.flush()
interval = random.uniform(2, 90)
print(unixtTime2StringTime(time.time()), cursor, interval)
result = cnfolMessageRequest(zjNet, rf, cursor)
print(result)
for status in result:
insertCusumError +=1
if status == record.insertStatus.Success:
insertCusumError=0
if insertCusumError == 20:
raise someError("insertCusumError")
if cursor == '2018-03-01':
raise someError("last date have data,had get all msg")
cursor = getPreviousDate(cursor)
time.sleep(interval)
def cnfolMessageRequest(net, rf, dateCursor='2018-03-01'):
net = network.net()
if net.loadCookie(__ZJ_MAINPAGE__) is None:
raise network.networkError("__ZJ_MAINPAGE__")
for i in range(__ERROR_RETRY__):
print(i, __ZJ_LIVE_ADD__)
resStr = net.getResponseData(__ZJ_LIVE_ADD__, {'livedate':dateCursor, 'order':1})
resList = rf.getHTMLObject(resStr)
if len(resList) > 0: # 请求成功跳出重试循环
break
if i == __ERROR_RETRY__:
raise someError("__ERROR_RETRY__:")
rows = []
for item in resList:
content = item[1]
createdTime = dateCursor + ' ' + item[0] + ':00'
row = {'contentHASH': rf.getMD5(content), 'id': '',
'source': __ZJ_TAG__, 'content': content,
'created_at': createdTime}
rows.append(row)
return rf.addARows(rows)
def wallstreetCollectorProcess(startTime=1552490575):
wsNet = network.net()
if wsNet.loadCookie(__WS_MAINPAGE__) is None:
raise network.networkError("__WS_MAINPAGE__")
rf = record.recordfile(__RECORDFILE_WS__, __FIELDNAMES__)
cursor = startTime
insertCusumError = 0
lockfile = open('./lockPage_ws','w+')
while cursor:
lockfile.seek(0)
lockfile.write(str(cursor))
lockfile.flush()
interval = random.uniform(2, 90)
print(unixtTime2StringTime(time.time()), cursor, interval)
result,nextCursor = wallstreetMessageRequest(wsNet, rf, cursor)
print(result)
for status in result:
insertCusumError +=1
if status == record.insertStatus.Success:
insertCusumError=0
if insertCusumError == 20:
raise someError("insertCusumError")
cursor = nextCursor
time.sleep(interval)
def __test__():
dataset = mnist_rotation.DataSet("mnist/train-images.idx3-ubyte",
"mnist/train-labels.idx1-ubyte",
"mnist/t10k-images.idx3-ubyte",
"mnist/t10k-labels.idx1-ubyte")
X = tf.placeholder(tf.uint8, [None, 42, 42, 1])
Y = tf.placeholder(tf.uint8, [None, 1])
images = (tf.cast(X, tf.float32) - 128.0) / 128.0
labels = tf.cast(Y, tf.int32)
stn, stn_params = network.net("stn", images, 6)
#stn = layers.tanh("stn_tanh",fcn0)
#stn
images_ = transformer.batch_transformer(images, stn, [42, 42])
sess = tf.Session()
stn_saver = tf.train.Saver(stn_params)
stn_saver.restore(sess, "./transformer_model_stn")
for i in range(1):
images, image_ = sess.run([images, images_],
feed_dict={
(X, Y): dataset.getTrainBatch(BATCH_SIZE)
})
images = images * 128 + 128
image_ = image_ * 128 + 128
for j in range(BATCH_SIZE):
cv2.imwrite(str(j) + "_src.jpg", images[j])
cv2.imwrite(str(j) + "_dst.jpg", image_[j])
def sinaCollectorProcess(startPage=1):
sinaNet = network.net()
if sinaNet.loadCookie(__SINA_MAINPAGE__) is None:
raise network.networkError("__SINA_MAINPAGE__")
rf = record.recordfile(__RECORDFILE_SINA__, __FIELDNAMES__)
page = startPage
insertCusumError = 0
lockfile = open('./lockPage_sina','w+')
while page:
lockfile.seek(0)
lockfile.write(str(page))
lockfile.flush()
interval = random.uniform(2, 90)
print(unixtTime2StringTime(time.time()), page, interval)
result = sinaMessageRequest(sinaNet, rf, page)
print(result)
for status in result:
insertCusumError +=1
if status == record.insertStatus.Success:
insertCusumError=0
if insertCusumError == 20 and len(sys.argv)<=2:
raise someError("insertCusumError")
if sys.argv[2] == 'test' and len(sys.argv) == 3:
if insertCusumError >= 10:
raise someError("test error: insertCusumError:", insertCusumError, "page:", page)
page += 1
time.sleep(interval)
def main(args):
# Create model directory for saving trained models
if not os.path.exists(args.model_path):
os.makedirs(args.model_path)
test_dataset = a2d_dataset.A2DDataset_test(test_cfg, args.dataset_path)
data_loader = DataLoader(test_dataset,
batch_size=1,
shuffle=False,
num_workers=1)
# define and load pre-trained model
model = net('se_resnext101')
model.cuda()
model.load_state_dict(
torch.load(
os.path.join(args.model_path,
'seresnext_mode_train_56.2_57.8.ckpt')))
results = np.zeros((data_loader.__len__(), args.num_cls))
model.eval()
# prediction and saving
with torch.no_grad():
for batch_idx, data in enumerate(data_loader):
# mini-batch
images = data.to(device)
output = model(images)
output = (
torch.nn.functional.sigmoid(output)).cpu().detach().numpy()
output[output >= 0.4] = 1
output[output < 0.4] = 0
results[batch_idx, :] = output
with open('results_netid.pkl', 'wb') as f:
pickle.dump(results, f)
def Choose_Model(self, Model_index):
# choose model structure
if self.model_name == 'Net':
from network import Net as net
elif self.model_name == 'Net_new3':
from network import Net_new3 as net
elif self.model_name == 'Net_new4':
from network import Net_new4 as net
else:
raise Exception('error: undefined Model_name')
self.model = net(num_channels=self.num_channels, scale_factor=self.scale_factor, d=32, s=5, m=1)
# choose pretrained model
if Model_index==0:
print('no need pretrained model')
# self.model.weight_init_fromMAT_s() #init from trained model
self.model.weights_init_kaiming()
elif Model_index==1:
self.model.load_state_dict(torch.load(self.pretrained_model))
print('\nTrained generator model is loaded:' + self.pretrained_model)
elif Model_index==2:
self.model.weight_init()
pretrained_dict = torch.load(self.pretrained_model, map_location=torch.device('cpu'))
model_dict = self.model.state_dict()
# 1. filter out unnecessary keys
pretrained_dict = {k: v for k, v in pretrained_dict.items() if k in model_dict}
# 2. overwrite entries in the existing state dict
model_dict.update(pretrained_dict)
self.model.load_state_dict(model_dict)
print('model partially initialized')
else:
raise Exception('error: undefined Model_index')
def main():
# parse arguments
args = parse_args()
if args is None:
exit()
if args.gpu_mode and not torch.cuda.is_available():
raise Exception("No GPU found, please run without --gpu_mode=False")
# print 'scale factor = ', scale_factor, \
# '\ntest_dir =', args.test_dataset,\
from network import Net_new4 as net
#from network import Net as net
model = net(num_channels=1, scale_factor=4, d=32, s=5, m=1)
#model = net(num_channels=1, scale_factor=4, d=32, s=5, m=5)
#model.load_state_dict(torch.load(args.pretrained_model, map_location = torch.device('cpu')))
#In GPU type
model.load_state_dict(torch.load(args.pretrained_model))
# for param_tensor in model.state_dict():
# #print(param_tensor)
# #print(model.state_dict()[param_tensor].size())
# if 'act' in param_tensor:
# print(param_tensor)
# print(model.state_dict()[param_tensor])
# print(model)
model_name_save = './1104_augument_based_1430_epoch110_3pm.onnx'
x = torch.randn(1, 1, 426, 570, requires_grad=False).type(torch.float)
torch_out = model(x)
torch.onnx.export(model, x, model_name_save, export_params=True)
def main():
# parse arguments
args = parse_args()
if args is None:
exit()
# if args.gpu_mode and not torch.cuda.is_available():
# raise Exception("No GPU found, please run without --gpu_mode=False")
# print 'scale factor = ', scale_factor, \
# '\ntest_dir =', args.test_dataset,\
from network import Net as net
#from network import Net as net
model = net(num_channels=1, scale_factor=4, d=32, s=5, m=1)
#model = net(num_channels=1, scale_factor=4, d=32, s=5, m=5)
#model.load_state_dict(torch.load(args.pretrained_model, map_location = torch.device('cpu')))
#In GPU type
model.load_state_dict(torch.load(args.pretrained_model,
map_location='cpu'))
# for param_tensor in model.state_dict():
# #print(param_tensor)
# #print(model.state_dict()[param_tensor].size())
# if 'act' in param_tensor:
# print(param_tensor)
# print(model.state_dict()[param_tensor])
# print(model)
x = torch.randn(1, 1, 1024, 768, requires_grad=False).type(torch.float)
stat(model, (1, 1024, 768))
def Choose_Model(self, Model_index):
# choose model structure
if self.model_name == 'Net':
from network import Net as net
elif self.model_name == 'Net_new3':
from network import Net_new3 as net
elif self.model_name == 'Net_new4':
from network import Net_new4 as net
elif self.model_name == 'FSRCNN':
from network import FSRCNN as net
elif self.model_name == 'FSRCNN_d1':
from network import FSRCNN_d1 as net
elif self.model_name == 'FSRCNN_d2':
from network import FSRCNN_d2 as net
elif self.model_name == 'FSRCNN_ps1':
net = importlib.import_module("model.FSRCNN_ps1").FSRCNN_ps1
elif self.model_name == 'FSRCNN_ps2_HN':
net = importlib.import_module("model.FSRCNN_ps2_HN").FSRCNN_ps2_HN
else:
raise Exception('error: undefined Model_name')
self.model = net(scale_factor=self.scale_factor,
num_channels=1,
d=56,
s=12,
m=4)
# choose pretrained model
if Model_index == 0:
print('no need pretrained model')
# self.model.weight_init_fromMAT_s() #init from trained model
#self.model.weights_init_kaiming()
elif Model_index == 1:
state_dict = torch.load(self.pretrained_model)
from collections import OrderedDict
new_state_dict = OrderedDict()
for k, v in state_dict.items():
#print(k)
name = k[7:] # remove `module.`
new_state_dict[name] = v
# load params
self.model.load_state_dict(new_state_dict)
#self.model.load_state_dict(torch.load(self.pretrained_model))
print('Trained generator model is loaded:', self.pretrained_model)
elif Model_index == 2:
self.model.weight_init()
pretrained_dict = torch.load(self.pretrained_model)
model_dict = self.model.state_dict()
# 1. filter out unnecessary keys
pretrained_dict = {
k: v
for k, v in pretrained_dict.items() if k in model_dict
}
# 2. overwrite entries in the existing state dict
model_dict.update(pretrained_dict)
self.model.load_state_dict(model_dict)
print('model partially initialized')
else:
raise Exception('error: undefined Model_index')
def __init__(self, data, alpha=0.000001):
if not isinstance(data, pd.DataFrame):
return None
else:
self.alpha = alpha
self.data = data.copy()
self.net = net(data=self.data)
self.net.calc_cpt(self.data)
self.scores = {}
def __init__(self, data, alpha=0.000001):
if not isinstance(data, pd.DataFrame):
return None
else:
self.data = data.copy()
self.net = net(data=self.data)
self.net.calc_cpt(self.data)
self.scores = {}
self.mi_weights = {}
self.alpha = alpha
print(self.net.nds.keys())
self.calc_norm_mi_Weights()
def generate_frames(first_frame_path_name, res_npy_ori_test_path,
res_npy_trs_test_path, model_path_name, gen_frames_path):
batch_shape = (1, img_height, img_width, 3)
with tf.Session() as sess:
X_content = tf.placeholder(tf.float32,
shape=batch_shape,
name="X_content")
Y_content = tf.placeholder(tf.float32,
shape=batch_shape,
name="Y_content")
preds = network.net((X_content + 255.0) / 255.0 / 2)
generated_frames_array = []
last_frame = rec.img_to_tensor(first_frame_path_name)
Y_content_flat = tf.reshape(Y_content,
[-1, img_width * img_height * 3])
preds_flat = tf.reshape(preds, [-1, img_width * img_height * 3])
loss = tf.reduce_mean(
tf.reduce_sum(tf.square(preds_flat - Y_content_flat),
reduction_indices=[1]))
tf.train.Saver().restore(sess, model_path_name)
res_ori_test = np.load(res_npy_ori_test_path)
res_trs_test = np.load(res_npy_trs_test_path)
# res_ori_test = rec.get_frames_tensors(frame_ori_path, 15001, 17982)
# res_trs_test = rec.get_frames_tensors(frame_trs_path, 15001, 17982)
X_batch = np.zeros(batch_shape, dtype=np.float32)
Y_batch = np.zeros(batch_shape, dtype=np.float32)
for i in range(int(len(res_ori_test))):
X_batch_array, Y_batch_array = our_utils.next_batch(
res_ori_test, res_trs_test, i)
X_batch[0] = X_batch_array[0]
Y_batch[0] = Y_batch_array[0]
to_get = [loss, preds]
test_feed_dict = {X_content: X_batch, Y_content: Y_batch}
tup = sess.run(to_get, feed_dict=test_feed_dict)
log.info("test " + str(i) + ", loss: " +
str(np.sqrt(tup[0] / img_width / img_height / 3)))
last_frame = last_frame + tup[1]
generated_frames_array.append(last_frame)
if not os.path.exists(gen_frames_path): os.makedirs(gen_frames_path)
for i in range(len(generated_frames_array)):
scipy.misc.imsave(gen_frames_path + 'frame_' + str(i) + '.jpg',
generated_frames_array[i][0])
def __init__(self,user_queue,out_queue,id,p_trigger,d_trigger,c_trigger):
QtCore.QThread.__init__(self,)
self.queue = user_queue
self.code_queue = out_queue
self.info = None
self.cToken = ''
self.captcha = ''
self.code = ''
self.timeStart = ''
self.timeSlotID = ''
self.hasStop = False
self.network = net()
self.id = id
self.p_trigger = p_trigger
self.d_trigger = d_trigger
self.c_trigger = c_trigger
def train(self, iterations=10, maxmiss=10):
"""
Train using Recommended Greedy Algorithm
"""
scores = {'Iteration': [], 'Network': [], 'Score': []}
score_check = maxmiss
niter = iterations
nodes = [i for i in self.data.columns]
best = score_pom(export_pom(self.net, by='label'), self.data)
#print("START LOOP")
while score_check > 0 and niter > 0:
n = net(data=self.data)
n.import_dag(self.net.export_dag())
ops = [n.add_edge, n.del_edge, n.rev_edge]
for f in ops:
# Choose the first node in a uniform, random way
v1 = np.random.choice(nodes)
# Choose the second with probabilities weighted by mi
v2 = np.random.choice(self.mi_weights[v1].index,
p=self.mi_weights[v1])
f(v1, v2)
if n.acyclic():
n.calc_cpt(self.data, alpha=self.alpha)
score = score_pom(export_pom(n, by='label'), self.data)
scores['Iteration'].append(iterations - niter)
scores['Network'].append(n)
scores['Score'].append(score)
#print(best, score, niter, score_check)
if score > best:
self.net = n
best = score
niter = niter - 1
score_check = maxmiss
continue
else:
score_check = score_check - 1
niter = niter - 1
continue
else:
niter = niter - 1
continue
self.scores = scores
def main(args):
# Create model directory for saving trained models
if not os.path.exists(args.model_path):
os.makedirs(args.model_path)
test_dataset = a2d_dataset.A2DDataset(train_cfg, args.dataset_path)
data_loader = DataLoader(test_dataset, batch_size=4, shuffle=True, num_workers=4) # you can make changes
# Define model, Loss, and optimizer
# model.to(device)
# model = net(num_classes=args.num_cls)
model = net(args.num_cls)
model.cuda()
criterion = nn.BCELoss()
optimizer = torch.optim.Adam(model.parameters(), lr=args.learning_rate)
# optimizer = torch.optim.SGD(model.parameters(), lr=args.learning_rate, momentum=args.momentum, weight_decay=args.weight_decay)
# Train the models
total_step = len(data_loader)
for epoch in range(args.num_epochs):
t1 = time.time()
for i, data in enumerate(data_loader):
# mini-batch
images = data[0].to(device)
labels = data[1].type(torch.FloatTensor).to(device)
#labels.squeeze(1)
# Forward, backward and optimize
outputs = model(images)
loss = criterion(outputs, labels)
model.zero_grad()
loss.backward()
optimizer.step()
# Log info
if i % args.log_step == 0:
print('Epoch [{}/{}], Step [{}/{}], Loss: {:.4f}'
.format(epoch, args.num_epochs, i, total_step, loss.item()))
# Save the model checkpoints
if (i + 1) % args.save_step == 0:
torch.save(model.state_dict(), os.path.join(
args.model_path, 'net.ckpt'))
t2 = time.time()
print(t2 - t1)
def __init__(self, parent = None):
self.healthFactor = random.randrange(15, 25)
self.speedFactor = random.randrange(80,120)
self.eyes = random.randrange(1, 1000)
self.color = (0, 0, 0)
self.eatingHabit = "Veg"
self.nn = network.net(2, 1)
if parent != None:
self.healthFactor = parent.dna.healthFactor
self.speedFactor = parent.dna.speedFactor
self.eyes = parent.dna.eyes
self.color = parent.dna.color
self.eatingHabit = parent.dna.eatingHabit
self.nn = parent.dna.nn
m = random.random()
if m < 0.2:
self._mutate(parent)
def wallstreetMessageRequest(net, rf, startTime):
net = network.net()
if net.loadCookie(__WS_MAINPAGE__) is None:
raise network.networkError("__WS_MAINPAGE__")
reqAddr = __WS_LIVE_ADD__ % (startTime)
for i in range(__ERROR_RETRY__):
print(i, reqAddr)
resStr = net.getResponseData(reqAddr)
resDic = rf.getJsonObject(resStr, None)
resCode = resDic["code"]
if resCode == 20000: # 请求成功跳出重试循环
break
if i == __ERROR_RETRY__:
raise someError("__ERROR_RETRY__:", resCode, resDic["message"])
items = resDic['data']['items']
if items is None:
print(resStr)
raise someError('NULL Content,maybe wrong cusor')
rows = []
for item in items:
if item['is_calendar']: # is_calendar时，本条记录不要
continue
content = item['content_text'].replace('\n','')
row = {'contentHASH': rf.getMD5(content), 'id': item['id'],
'source': __WS_TAG__, 'content': content,
'created_at': unixtTime2StringTime(item['display_time'])}
row['tag_name'] = item['score']
channels = item['channels']
if 'global-channel' in channels:
channels.remove('global-channel')
row['tag_name2'] = '|'.join(item['channels'])
rows.append(row)
nextCursor= resDic['data']['next_cursor']
return rf.addARows(rows), int(nextCursor)
def main(args):
# Create model directory for saving trained models
if not os.path.exists(args.model_path):
os.makedirs(args.model_path)
test_dataset = a2d_dataset.A2DDataset(
val_cfg, args.dataset_path) # changed to val_cfg from test_cfg
data_loader = DataLoader(test_dataset,
batch_size=1,
shuffle=False,
num_workers=1)
# define load your model here
# model = net(args.num_cls).cuda()
model = net(args.num_cls).to(device)
model.load_state_dict(torch.load(os.path.join(args.model_path,
'net.ckpt')))
X = np.zeros((data_loader.__len__(), args.num_cls))
Y = np.zeros((data_loader.__len__(), args.num_cls))
print(data_loader.__len__())
model.eval()
with torch.no_grad():
for batch_idx, data in enumerate(data_loader):
# mini-batch
images = data[0].to(device)
labels = data[1].type(torch.FloatTensor).to(device)
output = model(images).cpu().detach().numpy()
target = labels.cpu().detach().numpy()
output[output >= 0.5] = 1
output[output < 0.5] = 0
X[batch_idx, :] = output
Y[batch_idx, :] = target
P = Precision(X, Y)
R = Recall(X, Y)
F = F1(X, Y)
print('Precision: {:.1f} Recall: {:.1f} F1: {:.1f}'.format(
100 * P, 100 * R, 100 * F))
def evaluate_model(res_npy_ori_test_path, res_npy_trs_test_path,
model_path_name):
batch_shape = (1, img_height, img_width, 3)
with tf.Session() as sess:
X_content = tf.placeholder(tf.float32,
shape=batch_shape,
name="X_content")
Y_content = tf.placeholder(tf.float32,
shape=batch_shape,
name="Y_content")
preds = network.net(X_content / 255.0)
Y_content_flat = tf.reshape(Y_content,
[-1, img_width * img_height * 3])
preds_flat = tf.reshape(preds, [-1, img_width * img_height * 3])
loss = tf.reduce_mean(
tf.reduce_sum(tf.square(preds_flat - Y_content_flat),
reduction_indices=[1]))
tf.train.Saver().restore(sess, model_path_name)
res_ori_test = np.load(res_npy_ori_test_path)
res_trs_test = np.load(res_npy_trs_test_path)
X_batch = np.zeros(batch_shape, dtype=np.float32)
Y_batch = np.zeros(batch_shape, dtype=np.float32)
for i in range(0, int(len(res_ori_test))):
X_batch_array, Y_batch_array = our_utils.next_batch(
res_ori_test, res_trs_test, i)
X_batch[0] = X_batch_array[0]
Y_batch[0] = Y_batch_array[0]
to_get = [loss, preds]
test_feed_dict = {X_content: X_batch, Y_content: Y_batch}
tup = sess.run(to_get, feed_dict=test_feed_dict)
log.info("test " + str(i) + "average loss: " +
str(np.sqrt(tup[0] / img_width / img_height / 3)))
fs = np.int(1 / dt)
target[t_stim:] = lowpass_filter(np.random.randn(n_step_stim) * sigma, lp,
fs) #Low-pass filtered white noise
#training params
alpha = dt * 0.1
step = 50
train_start = int(np.round(start_stim / dt))
#Initialize network
net = network.net(N,
pNI,
mean_delays,
tref,
G=G,
p=p,
mean_GE=mean_GE,
mean_GI=mean_GI,
ITonic=ITonic,
tau=tau,
tr=tr,
td=td)
input_res = np.zeros((N_in, nt)) #Input to the reservoir
net.w_res = np.multiply(
np.random.normal(0, 1, (N, N_in)),
np.random.rand(N, N_in) < p_in) #Input weights (M in methods)
phase = np.random.uniform(
0, 1, N_in) * np.pi * 2 #Initial phase configuration of the oscillators
for inp_cell in range(N_in):
input_res[inp_cell, t_stim:t_stim + n_step_stim] = A * (np.sin(
2 * np.pi * osc_frequencies[inp_cell] *
images = dataset.test_data[:300].float()
label = dataset.test_labels[:300]
features = images.view(300, 784)
writer.add_embedding(features, metadata=label, label_img=images.unsqueeze(1))
running_loss = 0.0
for epoch in range(10): # loop over the dataset multiple times
for i, data in enumerate(trainloader, 0):
# get the inputs
inputs, labels = data
# zero the parameter gradients
optimizer.zero_grad()
# forward + backward + optimize
outputs = net(inputs)
loss = criterion(outputs, labels)
loss.backward()
optimizer.step()
running_loss += loss.item()
if i % PRINT_EACH == 0:
### Example of the network
dummy_input = torch.autograd.Variable(torch.rand(1, 3, 224, 224))
resnet18 = models.resnet18(False)
writer.add_graph(resnet18, (dummy_input, ))
print('[%d, %5d]' % (epoch + 1, i + 1))
print(' Train loss: %.3f' % (running_loss / PRINT_EACH))
writer.add_scalar('Train loss', (running_loss / PRINT_EACH),
p_inj = 1
p_in = 0.3
p_fb = 1
p_res = 0.3
gain_inj = 23
gain_fb = 0
gain_in = 0.1
input_Inets = np.zeros((nb_Inets * N_Inets, nt))
input_Inets[:, t_stim:t_stim + n_step_stim] = 1
Inets = [
network.net(N_Inets,
p_NI,
mean_delays,
Refractory,
G=G_Inets,
p=p_Inets,
mean_GE=GE_Inets,
mean_GI=GI_Inets,
ITonic=9,
mean_TauFall_I=TauI) for GI_Inets in GI_Inets_list
]
osc_range = [5, 7]
N_osc = 3
#training params
alpha = dt * 0.1
step = 50
train_start = int(np.round(start_stim / dt))
#PLot settings
data = pd.DataFrame({
'G1': G1,
'G2': G2,
'G3': G3,
'G4': G4,
'G5': G5,
'G6': G6,
'G7': G7,
'G8': G8,
'G9': G9,
'G10': G10
})
bnet = BN.from_samples(data)
bnet.plot()
plt.show()
n = net(data=data)
n.add_edge(0, 3)
n.add_edge(0, 4)
n.add_edge(1, 4)
n.add_edge(1, 3)
n.add_edge(1, 2)
n.add_edge(5, 6)
n.add_edge(5, 7)
n.calc_cpt(data)
samp = sample_net(n, 1000)
bnet = BN.from_samples(samp)
bnet.plot()
plt.show()
import logging
import json
from paddle_fl.split_learning import HostExecutor
from paddle_fl.split_learning.core.table import LocalTable
from paddle_fl.split_learning.core.reader import TmpReader
import network
logging.basicConfig(
format='%(asctime)s %(levelname)-8s [%(filename)s:%(lineno)d] %(message)s',
datefmt='%Y-%m-%d %H:%M',
level=logging.DEBUG)
_LOGGER = logging.getLogger(__name__)
if __name__ == "__main__":
data_path = sys.argv[1]
host_input, consumer_input, label, prediction, cost = network.net()
main_program = fluid.default_main_program()
startup_program = fluid.default_startup_program()
place = fluid.CPUPlace()
table = LocalTable(data_path)
reader = TmpReader(place)
exe = HostExecutor(place, table, reader)
exe.load_program_from_full_network(
startup_program=startup_program,
main_program=main_program)
exe.start(7858)
def __train__():
#data
dataset = mnist_rotation.DataSet("mnist/train-images.idx3-ubyte",
"mnist/train-labels.idx1-ubyte",
"mnist/t10k-images.idx3-ubyte",
"mnist/t10k-labels.idx1-ubyte")
X = tf.placeholder(tf.uint8, [None, 42, 42, 1])
Y = tf.placeholder(tf.uint8, [None, 1])
images = (tf.cast(X, tf.float32) - 128.0) / 128.0
labels = tf.cast(Y, tf.int32)
stn, stn_params = network.net("stn", images, 6)
#!!!!!!!!!!!!!use tanh as activation
#stn_ = layers.tanh("stn_tanh",stn)
#stn
images_ = transformer.batch_transformer(images, stn, [42, 42])
"""
R = tf.placeholder(tf.uint8,[None,42,42,1])
RL = tf.placeholder(tf.uint8,[None,1])
R_ = (tf.cast(R,tf.float32) - 128.0)/128.0
RL_ = tf.cast(RL,tf.int32)
IMAGE = tf.concat([images_,R_],0)
LABEL = tf.concat([labels,RL_],0)
"""
#cnn
cnn, cnn_params = network.net("cnn", images_, CLASS_SIZE, trainable=True)
loss, cross_loss = layers.crossentropy("stn_loss", cnn, labels, BATCH_SIZE,
CLASS_SIZE)
global_step = tf.Variable(0, trainable=False)
opt = optimize(loss, global_step, 60000, BATCH_SIZE,
stn_params + cnn_params)
#cnn
"""_
cnn_,cnn_params_ = network.net("cnn",images,CLASS_SIZE,trainable = True,reuse = True)
loss_,cross_loss_ = layers.crossentropy("cnn_loss",cnn_,labels,BATCH_SIZE,CLASS_SIZE)
global_step_ = tf.Variable(0, trainable=False)
opt_ = optimize(loss_,global_step_,30000,BATCH_SIZE,cnn_params_)
#softmax = layers.softmax("cnn_softmax",cnn)
"""
"""
CNN, CNN_PARAMS = network.net("cnn",IMAGE,CLASS_SIZE,trainable = True,reuse = True:)
LOSS,CROSS_LOSS = layers.crossentropy("LOSS",CNN,LABEL,BATCH_SIZE*2,CLASS_SIZE)
GLOBAL_STEP = tf.Variable(0, trainable=False)
OPT = optimize(LOSS,GLOBAL_STEP,60000,BATCH_SIZE*2)
"""
sess = tf.Session()
init = tf.global_variables_initializer()
sess.run(init)
stn_saver = tf.train.Saver(stn_params)
cnn_saver = tf.train.Saver(cnn_params)
save_file_name = "./transformer_model"
for step in range(100):
#print "step = ",step
value = 0
global_step = step
start_time = time.time()
for epoch in range(int(60000 / BATCH_SIZE)):
#print epoch
#_cnn,_y = sess.run([cnn,Y],feed_dict = {(X,Y):dataset.getTrainBatch(BATCH_SIZE)})
#print _cnn,_y
opt_, loss_ = sess.run([opt, cross_loss],
feed_dict={
(X, Y):
dataset.getTrainBatch(BATCH_SIZE)
})
#value += _loss_value
print "current loss : ", loss_
#_opt_,_loss_value_ = sess.run([opt_, cross_loss_],feed_dict = {(X,Y):dataset.getTrainBatch(BATCH_SIZE,"TEST")})
#value += _loss_value
#print "current loss : ",_loss_value," ",_loss_value_
#_opt,_loss_value = sess.run([OPT , CROSS_LOSS],feed_dict = {(X,Y,R,RL):(dataset.getTrainBatch(BATCH_SIZE),dataset.getTrainBatch(BATCH_SIZE))})
#value += _loss_value
#print _loss_value_
print("step = ", step)
print("loss = ", value / (60000 / BATCH_SIZE))
print("duration = ", time.time() - start_time)
#save model
if (step % 1 == 0):
save_path = stn_saver.save(sess, save_file_name + "_stn")
print(save_path)
save_path = cnn_saver.save(sess, save_file_name + "_cnn")
print(save_path)
shorten_factor=shorten_factor)
len_mel = mel_spec.shape[1]
#Resample spectrogram to match the number of time-steps of the simulation
rs_audio = len_output / len_mel
target = np.full((N_out, nt), np.min(mel_spec))
target[:, t_stim:
-len_buffer] = mel_spec[:,
np.round(np.linspace(0, len_mel - 1, len_output)
).astype(np.int)]
#Initialize network
net = network.net(N,
pNI,
mean_delays,
tref,
G=G,
p=p,
mean_GE=mean_GE,
mean_GI=mean_GI,
ITonic=ITonic)
net.tr = tr
net.td = td
input_res = np.zeros((N_in, nt))
N_in_net = int(np.round(p_in * N))
net.w_res = np.abs(
np.multiply(np.random.normal(0, 1, (N, N_in)),
np.random.rand(N, N_in) < p_in))
phase = np.random.uniform(0, 1, N_in) * np.pi
for inp_cell in range(N_in):
input_res[inp_cell, t_stim:t_stim + n_step_stim] = A * (
def main(args):
# Create model directory for saving trained models
if not os.path.exists(args.model_path):
os.makedirs(args.model_path)
test_dataset = a2d_dataset.A2DDataset(train_cfg, args.dataset_path)
data_loader = DataLoader(test_dataset,
batch_size=args.batch_size,
shuffle=True,
num_workers=12) # you can make changes
# Define model, Loss, and optimizer
model = net('se_resnext101')
model.cuda()
model.load_state_dict(
torch.load(os.path.join(args.model_path, 'net_F.ckpt')))
# for i, param in model.backbone.features.named_parameters():
# if 'stage4' or 'stage3' or 'stage2' or 'stage1'in i:
# print(i)
# param.requires_grad = True
# else:
# param.requires_grad = False
criterion = nn.BCEWithLogitsLoss()
# print(list(filter(lambda p: p.requires_grad, model.parameters())))
base_optimizer = optim.SGD(model.parameters(), lr=0.005, momentum=0.9)
# base_optimizer = optim.SGD(list(filter(lambda p: p.requires_grad, model.parameters())), lr=0.005, momentum=0.9)
optimizer = SWA(base_optimizer, swa_start=1, swa_freq=5, swa_lr=0.005)
# optimizer = optim.SGD(model.parameters(), lr=0.05, momentum=0.9)
# optimizer = optim.Adam(model.parameters(), lr=0.005, weight_decay=0.00005)
# scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(optimizer, args.num_epochs, eta_min=5e-5, last_epoch=-1)
scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer,
[5, 10, 20, 40, 75],
gamma=0.25)
model, optimizer = amp.initialize(model,
optimizer,
opt_level="O1",
verbosity=0)
# Train the models
total_step = len(data_loader)
best_P, best_R, best_F = 0, 0, 0
for epoch in range(args.num_epochs):
# scheduler.step()
print('epoch:{}, lr:{}'.format(epoch, scheduler.get_lr()[0]))
t1 = time.time()
for i, data in enumerate(data_loader):
optimizer.zero_grad()
# mini-batch
images = data[0].to(device)
labels = data[1].type(torch.FloatTensor).to(device)
# Forward, backward and optimize
outputs = model(images)
loss = criterion(outputs, labels)
with amp.scale_loss(loss, optimizer) as scaled_loss:
scaled_loss.backward()
if (i + 1) % 1 == 0:
torch.nn.utils.clip_grad_norm_(model.parameters(),
max_norm=5.0,
norm_type=2)
optimizer.step()
optimizer.zero_grad()
# optimizer.swap_swa_sgd()
# scheduler.step()
# Log info
if i % args.log_step == 0:
print('Epoch [{}/{}], Step [{}/{}], Loss: {:.4f}'.format(
epoch, args.num_epochs, i, total_step, loss.item()))
optimizer.swap_swa_sgd()
# Save the model checkpoints
# if (i + 1) % args.save_step == 0:
# torch.save(model.state_dict(), os.path.join(
# args.model_path, 'net.ckpt'))
scheduler.step()
t2 = time.time()
print('Time Spend per epoch: ', t2 - t1)
if epoch > -1:
val_dataset = a2d_dataset.A2DDataset(val_cfg, args.dataset_path)
val_data_loader = DataLoader(val_dataset,
batch_size=1,
shuffle=False,
num_workers=1)
X = np.zeros((val_data_loader.__len__(), 43))
Y = np.zeros((val_data_loader.__len__(), 43))
model.eval()
with torch.no_grad():
for batch_idx, data in enumerate(val_data_loader):
# mini-batch
images = data[0].to(device)
labels = data[1].type(torch.FloatTensor).to(device)
# output = model(images).cpu().detach().numpy()
output = model(images)
output = (torch.nn.functional.sigmoid(output)
).cpu().detach().numpy()
target = labels.cpu().detach().numpy()
output[output >= 0.5] = 1
output[output < 0.5] = 0
X[batch_idx, :] = output
Y[batch_idx, :] = target
P = Precision(X, Y)
R = Recall(X, Y)
F = F1(X, Y)
print('Precision: {:.1f} Recall: {:.1f} F1: {:.1f}'.format(
100 * P, 100 * R, 100 * F))
if (P > best_P):
torch.save(model.state_dict(),
os.path.join(args.model_path, 'net_P.ckpt'))
best_P = P
if (R > best_R):
torch.save(model.state_dict(),
os.path.join(args.model_path, 'net_R.ckpt'))
best_R = R
if (F > best_F):
torch.save(model.state_dict(),
os.path.join(args.model_path, 'net_F.ckpt'))
best_F = F
data = {} #Output dictionary used by figure2b.py
data['nets'] = []
cutoff = 0.95 #Cut-off correlation between the activity on each trial to consider a network as stable
print('Starting simulation.')
for net_i in range(nb_networks):
input_network = np.zeros(nt)
N_in = 1
unstable = True
mean_TauFall_I = mean_TauFall_I_list[net_i]
while unstable:
total_iEx = np.zeros((nb_epochs,nt))
total_iIn = np.zeros((nb_epochs,nt))
total_iInp = np.zeros((nb_epochs,nt))
spikes = np.zeros((N,nt,nb_epochs))
#Exc cells
net = network.net(N,pNI,mean_delays,tref,G=G,p=p, mean_GE = mean_GE, mean_GI = mean_GI, ITonic=ITonic, mean_TauFall_I=mean_TauFall_I,GaussSD=GaussSD)
N_in_net = p_in*net.NE
temp_w_in = np.abs(np.multiply(np.random.normal(0,1,(N_in_net,N_in)),np.random.rand(N_in_net,N_in)<p_in))
w_res = np.zeros((N,N_in))
w_res[net.E_idx,:] = temp_w_in #Only the excitatory units receive the step input
net.w_res = w_res
input_network[t_stim:e_stim] = A
for ep_i in range(nb_epochs):
#Simulation variables
gEx = np.zeros(N) #Conductance of excitatory neurons
gIn = np.zeros(N) #Conductance of excitatory neurons
F = np.full(N,np.nan) #Last spike times of each inhibitory cells
V = np.random.normal(net.mean_VRest,abs(0.02*net.mean_VRest),N) #Set initial voltage Exc
for t in range(nt):
import paddle.fluid as fluid
import numpy as np
import time
import grpc
import json
import yaml
import paddle
import network
import utils
if __name__ == "__main__":
paddle.enable_static()
place = fluid.CPUPlace()
exe = fluid.Executor(place)
slot1, slot2, label, prediction, cost = network.net()
main_program = fluid.default_main_program()
startup_program = fluid.default_startup_program()
#startup_program, main_program = split_nn.util.load_whole_program("whole_program")
exe.run(startup_program)
"""
fluid.io.load_persistables(
executor=exe,
dirname="whole_program/persistables_vars",
main_program=main_program)
"""
for i, item in enumerate(utils.data_iter("../data/input.json")):
uid, x1, x2, label = item
data = {