def __init__(self, name, n_actions, n_features, hidsizes, lr, ac_fn, vf_coef=0.1, seed=0):
np.random.seed(seed)
tf.set_random_seed(seed)
self.logger = get_logger(name)
self.n_actions = n_actions
self.n_features = n_features
self.hidsizes = hidsizes
self.ac_fn = ac_fn
self.lr = lr
self.vf_coef = vf_coef
self._build_net()
config = tf.ConfigProto(allow_soft_placement=True, log_device_placement=False)
config.gpu_options.allow_growth = True
self.sess = tf.Session(config=config)
self.sess.run(tf.global_variables_initializer())
from concurrent.futures import ProcessPoolExecutor
from common import util
LOGGER = util.get_logger()
MAX_THREADS = 10
TEST_URL = 'https://lpdaac.usgs.gov/'
def _sum(*args):
LOGGER.info(f'adding {args[0]} ...')
return sum(*args)
def main():
sum_inputs = [(1, 2), (3, 4)]
all_sums = []
with ProcessPoolExecutor() as pool:
# the process pool creates N number of Python interp processes
# where Ni s the number of available CPUs
# parallel calculation by mapping
for s in pool.map(_sum, sum_inputs):
all_sums.append(s)
# sum of the sums
total = sum(all_sums)
LOGGER.info(f'total = {total}')
if __name__ == '__main__':
with timed('loading data ...'):
if args.model == 'cnn':
X_train, y_train = get_data_and_labels('data/train.txt', False)
X_test, y_test = get_data_and_labels('data/test.txt', False)
else:
X_train = np.loadtxt('data/hog_X_train.csv', delimiter=',')
y_train = np.expand_dims(
np.loadtxt('data/hog_y_train.csv', delimiter=','), -1)
X_test = np.loadtxt('data/hog_X_test.csv', delimiter=',')
y_test = np.expand_dims(
np.loadtxt('data/hog_y_test.csv', delimiter=','), -1)
if args.model == 'lr':
from baselines.logistic_regression import LogisticRegression
logger = get_logger('lr_langevin' if args.langevin else 'lr')
model = LogisticRegression(
learning_rate=args.lr,
total_epoches=args.total_epoches,
langevin=args.langevin,
seed=args.seed,
logger=logger,
X_test=X_test,
y_test=y_test,
)
# from sklearn.linear_model.logistic import LogisticRegression
# model = LogisticRegression()
elif args.model == 'svm':
from baselines.svm import SVM
logger = get_logger(f'svm_{args.kernel}')
#环境检测
EnvCheck()
#获取host port
try:
host = sys.argv[1]
port = int(sys.argv[2])
except IndexError, e:
host = '0.0.0.0'
port = setting.TcpPort
#addr
g.addr = (host, port)
#队列
g.events_queue = Queue.Queue()
#日志
g.logger = util.get_logger(setting.logFileName)
#线程锁
g.condLock = threading.Condition()
#信号处理
regSignal()
#记录pid
logPid(setting.runPidFile)
#修改进程名称
util.setProcName("AutoDeployServer")
#db记录目录变化
daemonDoSomeThing(watchDirByDb)
from rl.a2c.a2c import Model
from rl.a2c.runner import Runner
from rl.common import set_global_seeds
from baselines.stochastic import Model as Stochastic
from baselines.rule import Model as Rule
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
if __name__ == '__main__':
set_global_seeds(args.seed)
logger = get_logger('trading')
logger.info(str(args))
env = Env('train')
# Instantiate the model objects (that creates defender_model and adversary_model)
model = Model(
ob_size=env.ob_size,
act_size=env.act_size,
learning_rate=args.lr,
latents=args.latents,
activation=args.activation,
optimizer=args.optimizer,
vf_coef=args.vf_coef,
ent_coef=args.ent_coef,
max_grad_norm=args.max_grad_norm
def __init__(
self,
name, # name of this model
env, # environment
latents, # network hidden layer sizes
lr=1e-5, # learning rate
activation='relu', # activation function
optimizer='adam', # optimization function
vf_coef=0.1, # vf_loss weight
ent_coef=0.01, # ent_loss weight
max_grad_norm=0.5): # how frequently the logs are printed out
sess = tf_util.get_session()
# output to both file and console
logger = get_logger(name)
output = logger.info
output(args)
activation = tf_util.get_activation(activation)
optimizer = tf_util.get_optimizer(optimizer)
# lr = tf.train.polynomial_decay(
# learning_rate=lr,
# global_step=tf.train.get_or_create_global_step(),
# decay_steps=total_epoches,
# end_learning_rate=lr/10,
# )
ob_size = env.ob_size
act_size = env.act_size
n_actions = env.n_actions
# placeholders for use
X = tf.placeholder(tf.float32, [None, ob_size], 'observation')
A = tf.placeholder(tf.int32, [None, act_size], 'action')
ADV = tf.placeholder(tf.float32, [None], 'advantage')
R = tf.placeholder(tf.float32, [None], 'reward')
with tf.variable_scope(name):
policy = build_policy(
observations=X,
act_size=act_size,
n_actions=n_actions,
latents=latents,
vf_latents=latents,
activation=activation
)
# Calculate the loss
# Total loss = Policy gradient loss - entropy * entropy coefficient + Value coefficient * value loss
# Policy loss
neglogpac = policy.neglogp(A)
# L = A(s,a) * -logpi(a|s)
pg_loss = tf.reduce_mean(ADV * neglogpac)
# Entropy is used to improve exploration by limiting the premature convergence to suboptimal policy.
entropy = tf.reduce_mean(policy.entropy())
# Value loss
vf_loss = losses.mean_squared_error(tf.squeeze(policy.vf), R)
loss = pg_loss - entropy * ent_coef + vf_loss * vf_coef
# gradients and optimizer
params = find_trainable_variables(name)
grads = tf.gradients(loss, params)
if max_grad_norm is not None:
# Clip the gradients (normalize)
grads, grad_norm = tf.clip_by_global_norm(grads, max_grad_norm)
grads = list(zip(grads, params))
# 3. Make op for one policy and value update step of A2C
trainer = optimizer(learning_rate=lr)
_train = trainer.apply_gradients(grads)
# Add ops to save and restore all the variables.
saver = tf.train.Saver(tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, scope=name))
def step(obs):
action, value = sess.run([policy.action, policy.vf], feed_dict={
X: np.reshape(obs, (-1, ob_size))
})
return action, value
def value(obs):
return sess.run(policy.vf, feed_dict={
X: np.reshape(obs, (-1, ob_size))
})
def train(obs, rewards, actions, values):
# Here we calculate advantage A(s,a) = R + yV(s') - V(s)
# rewards = R + yV(s')
advs = rewards - values
td_map = {X:obs, A:actions, ADV:advs, R:rewards}
policy_loss, value_loss, policy_entropy, _ = sess.run(
[pg_loss, vf_loss, entropy, _train],
td_map
)
return policy_loss, value_loss, policy_entropy
def save(save_path):
saver.save(sess, save_path)
print(f'Model saved to {save_path}')
def load(load_path):
saver.restore(sess, load_path)
print(f'Model restored from {load_path}')
self.train = train
self.step = step
self.value = value
self.output = output
self.save = save
self.load = load
tf.global_variables_initializer().run(session=sess)
def __init__(self, mosaic, mosaicConf, utilsConf):
"""
Calculate the column density of a mosaic (applying corrections) for the following species:
'HI'(default), and 'HISA','HI+HISA','HI+CO','HI+HISA+CO' only for CGPS/SGPS
"""
self.survey = mosaic.survey
self.mosaic = mosaic.mosaic
self.species = mosaic.newspec # specified by the user
self.type = mosaic.type
self.logger = get_logger(self.survey + '_' + self.mosaic + '_' + self.species + '_ColumnDensity')
file = ''
path = ''
flag = ''
sur = self.survey.lower()
if self.species == 'HI':
path = get_path('lustre_' + sur + '_hi_column_density')
flag = 'HI_unabsorbed'
if sur == 'lab':
flag = 'HI'
elif self.species == 'HISA':
path = get_path('lustre_' + sur + '_hisa_column_density')
flag = 'HISA'
elif self.species == 'HI+HISA':
path = get_path('lustre_' + sur + '_hi_column_density')
flag = 'HI+HISA'
elif self.species == 'CO':
path = get_path('lustre_' + sur + '_co_column_density')
flag = 'H2'
elif self.species == 'WCO':
path = get_path('lustre_' + sur + '_co_column_density')
flag = 'H2'
elif self.species == 'HI+CO':
path = get_path('lustre_' + sur + '_hi_column_density')
flag = 'HI+H2'
elif self.species == 'HI+HISA+CO':
path = get_path('lustre_' + sur + '_hi_column_density')
flag = 'HI+HISA+H2'
file = path + self.survey + '_' + self.mosaic + '_' + flag + '_column_density.fits'
check_for_files([file], existence=True)
self.logger.info("Open file and get data...")
# Array to store results
N = np.zeros((mosaic.ny, mosaic.nx), dtype=float)
Ts = float(utilsConf['tspin']) # [Excitation (or Spin) Temperature] = K (150)
Tbg = float(utilsConf['tcmb']) # [Cosmic Microwave Background (CMB)] = K
if not self.species == 'WCO':
dv = np.fabs(mosaic.dz / 1000.) # [velocity] = km s-1
C = float(utilsConf['c']) # [costant] = cm-2
# Corrections for latitude (almost negligible)
cosdec = np.cos(np.radians(mosaic.yarray)) # [lat. correction] = rad
if self.survey == 'Galprop':
if self.species == 'WCO':
# Get emission data and velocity interval
wco = np.sum(mosaic.observation, axis=0)
# 2 accounts for the convertion from molecules to atoms
N = 2 * wco * float(utilsConf['xfactor']) * cosdec
elif self.survey == 'LAB':
# Get data
Tb = mosaic.observation[:, :, :]
del mosaic.observation
# Tb must be < Ts, otherwise problems with log
index = np.where(Tb >= Ts)
Tb[index] = 0.999 * Ts
# Optical depth correction
# Without the continuum component
cTb = np.log((Ts) / (Ts - Tb)) * Ts
# cTb = -log(1-(Tb/(Ts-Tbg))) * Ts
# Integrated brightness temperature over velocity (axis = 0)
ITb = np.sum(cTb, axis=0)
self.logger.info("Calculating NHI...")
# Column density
N = C * ITb * dv # [N] = cm-2
# Corrected column density
N = N * cosdec
elif self.survey == 'Dame':
if self.species == 'WCO':
# Get emission data and velocity interval
wco = mosaic.observation
del mosaic.observation
# 2 accounts for the convertion from molecules to atoms
N = 2 * wco * float(utilsConf['xfactor']) * cosdec
elif self.survey == 'CGPS' or self.survey == 'SGPS':
# Used to skip calculation (see below) - do not change it!
flagHI, flagCO, flagHISA = True, True, True
# If the two column density files (or one of them) already exist,
# than add them up and skip the calculation (or skip the unnecessary one)
if self.species == 'HI+HISA':
path1 = get_path('lustre_' + sur + '_hi_column_density')
file1 = path1 + self.survey + '_' + self.mosaic + '_HI_unabsorbed_column_density.fits'
path2 = get_path('lustre_' + sur + '_hisa_column_density')
file2 = path2 + self.survey + '_' + self.mosaic + '_HISA_column_density.fits'
N, flagHI, flagHISA = checkToGetData(self.logger, file1, file2)
if self.species == 'HI+CO':
path1 = get_path('lustre_' + sur + '_hi_column_density')
file1 = path1 + self.survey + '_' + self.mosaic + '_HI_unabsorbed_column_density.fits'
path2 = get_path('lustre_' + sur + '_co_column_density')
file2 = path2 + self.survey + '_' + self.mosaic + '_H2_column_density.fits'
N, flagHI, flagCO = checkToGetData(self.logger, file1, file2)
if self.species == 'HI+HISA+CO':
path1 = get_path('lustre_' + sur + '_hi_column_density')
file1 = path1 + self.survey + '_' + self.mosaic + '_HI_unabsorbed_column_density.fits'
path2 = get_path('lustre_' + sur + '_hisa_column_density')
file2 = path2 + self.survey + '_' + self.mosaic + '_HISA_column_density.fits'
path3 = get_path('lustre_' + sur + '_co_column_density')
file3 = path3 + self.survey + '_' + self.mosaic + '_H2_column_density.fits'
N, flagHI, flagHISA, flagCO = checkToGetData(self.logger, file1, file2, file3)
# Computing Column Density
if self.species != 'CO':
# Get HI emission data
Tb = mosaic.observation[0, :, :, :]
# Setting the negative/0-values to Tcmb
# Tb = where( (Tb<0.) | (Tb==0.),Tbg,Tb)
# HI Column Density
if self.species != 'HISA' and flagHI:
NHI = np.zeros((mosaic.ny, mosaic.nx), dtype=float)
self.logger.info("Initializing parameters...")
self.logger.info("1) Ts = %.2f K" % Ts)
self.logger.info("2) dV = %.2f km/s" % dv)
self.logger.info("3) Tb(min) = %.2f K, Tb(max) = %.2f K" % (np.amin(Tb), np.amax(Tb)))
self.logger.info("Calculating NHI...")
# Optical depth correction
# With the continuum component
# Tfunc = (Ts-Tc)/(Ts-Tc-Tb)
# Without the continuum component
Tfunc = Ts / (Ts - Tb)
Tfunc[Tfunc < 1.] = 1. # <------ TO JUSTIFY
cTb = np.log(Tfunc) * Ts
# Integrated brightness temperature over velocity (axis = 0)
ITb = np.sum(cTb[mosaic.zmin:mosaic.zmax, :, :], axis=0)
# Column density
NHI = C * ITb * dv # [NHI] = cm-2
N = NHI * cosdec
# HISA Column Density
if self.species not in ['HI', 'HI+CO'] and flagHISA:
NHISA = np.zeros(mosaic.observation.shape, dtype=np.float)
# Get HI continuum data
pathc = get_path(sur + '_hi_continuum')
continuum = pathc + self.survey + '_' + self.mosaic + '_1420_MHz_I_image.fits'
check_for_files([continuum])
data, header = fits.getdata(continuum, 0, header=True)
data[np.isnan(data)] = 0.
Tc = data
if self.survey == 'CGPS':
Tc = data[0, 0, :, :]
if self.survey == 'SGPS':
Tc = data[:, :]
# Tc = where( (Tc<0.) | (Tc==0.),Tbg,Tc)
# Get HISA data
path_hisa_dat = get_path(sur + '_hisa_dat')
amplitude = path_hisa_dat + self.survey + '_' + self.mosaic + '_HISA.dat'
check_for_files([amplitude])
input = open(amplitude, 'r')
lines = input.readlines()
self.logger.info("Initializing parameters...")
self.logger.info("1) dv = %.2f km/s" % dv)
self.logger.info("2) Tb(min) = %.2f K, Tb(max) = %.2f K" % (np.amin(Tb), np.amax(Tb)))
self.logger.info("3) Tc(min) = %.2f K, Tc(max) = %.2f K" % (np.amin(Tc), np.amax(Tc)))
galax_dist = open("galaxy_" + self.mosaic + "_" + self.species + ".dat", "w")
string = ""
for line in lines:
na = float(line.split('\n')[0].split()[0]) # HISA region
nb = float(line.split('\n')[0].split()[1]) # HISA location
Tu = float(line.split('\n')[0].split()[2]) # Unabsorbed brightness (Tu)
dT = float(line.split('\n')[0].split()[3]) # Delta T (always negative)
# print "%.2f\t%.2f\t%.2f\t%.2f"%(na,nb,nc,nd)
m = np.floor(nb / 1024 / 1024)
ha = nb - m * 1024 * 1024
l = np.floor(ha / 1024)
k = ha - l * 1024
# Params
glon = mosaic.xarray[k]
glat = mosaic.yarray[l] # <--- 160? (it's in the martin's idl code)
vlsr = mosaic.zarray[m] / 1000.
d = rotCurveMPohl(self.logger, glon, glat, vlsr) # kpc
string = "%s %s %s %s\n" % (d, glon, glat, vlsr)
galax_dist.write(string)
theta = np.radians(mosaic.dy) # rad
ds = d * np.tan(theta) * 1e3 # pc
A1 = float(utilsConf['pc2cm']) * float(utilsConf['poverk'])
A2 = float(utilsConf['fn']) * ds / (C * dv)
A = A1 * A2
B = Tc[l, k] + float(utilsConf['p']) * Tu
init = [1., 10.]
def equations(xx):
'''
Ts and tau functions - S.J.Gibson et al
(The Astrophysical Journal, 540:852-862, 2000)
'''
tt, T = xx
# Equation (6)
Tfunc = (T - B) / (T - B - dT)
if Tfunc < 1.:
Tfunc = 1. # Tbg # <------ TO JUSTIFY
f1 = np.log(Tfunc) - tt
# Equation (9)
ttfunc = A / tt
if ttfunc < 0.:
ttfunc = 0. # <------ TO JUSTIFY
f2 = np.sqrt(ttfunc) - T
return np.array([f1, f2], dtype=float)
(tau, Ts), infodict, ier, mesg = fsolve(equations, init, full_output=1)
# plotFunc(tau,Ts)
TsMin = Tbg
if Ts < TsMin:
Ts = TsMin
Tfunc = (Ts - B) / (Ts - B - dT)
tau = np.log(Tfunc)
TsMax = Tc[l, k] + (dT + Tu) + (float(utilsConf['p']) - 1.) * Tu
if Ts > TsMax:
# For Ts = TsMax, tau --> +oo
Ts = TsMax
tau = 1e4
if tau < 0.:
Ts = 0.
tau = np.log(B / (B + dT))
# print Ts,tau
solution = False
if ier == 1:
solution = True
NHISA[0, m, l, k] = tau * Ts * dv * C
NHISA[0, m, l, k] = NHISA[0, m, l, k] * np.cos(np.radians(mosaic.yarray[l]))
if not solution:
# print "Could not find a valid solution:"
# print mesg
NHISA[0, m, l, k] = 0.
galax_dist.close()
self.logger.info("Calculating NHI...")
# Corrected column density
N = N + np.sum(NHISA[0, mosaic.zmin:mosaic.zmax, :, :], axis=0)
if self.species in ['CO', 'HI+CO', 'HI+HISA+CO'] and flagCO:
# Get CO emission data
pathco = get_path('lustre_' + sur + '_co')
co = pathco + self.survey + '_' + self.mosaic + '_CO_line.fits'
check_for_files([co])
Wco, header = fits.getdata(co, 0, header=True)
# 2 accounts for the convertion from molecules to atoms
N = N + 2 * Wco * float(utilsConf['xfactor']) * cosdec
if self.species == 'CO':
kmin = 30
kmax = 230
# Get emission data and velocity interval
Tb = mosaic.observation[0, :, :, :]
wco = np.zeros((mosaic.ny, mosaic.nx), dtype=np.float)
wco = np.sum(Tb[kmin:kmax, :, :], axis=0) * dv
# 2 accounts for the convertion from molecules to atoms
N = 2 * wco * float(utilsConf['xfactor']) * cosdec
# Store results
newheader = fits.Header()
newheader["ctype1"] = ("GLON-CAR", "Coordinate type")
newheader["crval1"] = (mosaic.keyword["crval1"], "Galactic longitude of reference pixel")
newheader["crpix1"] = (mosaic.keyword["crpix1"], "Reference pixel of lon")
newheader["cdelt1"] = (mosaic.keyword["cdelt1"], "Longitude increment")
newheader["crota1"] = (mosaic.keyword["crota1"], "Longitude rotation")
newheader["cunit1"] = ("deg", "Unit type")
newheader["ctype2"] = ("GLAT-CAR", "Coordinate type")
newheader["crval2"] = (mosaic.keyword["crval2"], "Galactic latitude of reference pixel")
newheader["crpix2"] = (mosaic.keyword["crpix2"], "Reference pixel of lat")
newheader["cdelt2"] = (mosaic.keyword["cdelt2"], "Latitude increment")
newheader["crota2"] = (mosaic.keyword["crota2"], "Latitude rotation")
newheader["cunit2"] = ("deg", "Unit type")
newheader["bunit"] = ("atoms cm-2", "Map units")
newheader["datamin"] = ("%e" % np.amin(N), "Min value")
newheader["datamax"] = ("%e" % np.amax(N), "Max value")
newheader["object"] = ("Mosaic " + self.mosaic, self.survey + " Mosaic")
results = fits.PrimaryHDU(N, newheader)
# Output file
self.logger.info('Write data to a fits file in...')
results.writeto(file, output_verify='fix')
self.logger.info('{}'.format(path))
self.logger.info('Done')
