def __init__(self,
name,
snvs_valid,
snvs_invalid,
indels_valid,
indels_invalid,
sensitivity=None,
precision=None):
self.name = name
self.sensitivity = random.uniform(
0.5, 0.95) if not sensitivity else sensitivity
self.precision = random.uniform(0.5,
0.95) if not precision else precision
goodfrac = self.sensitivity
badfrac = self.sensitivity * (1. - self.precision) / self.precision
self.snvs = random.sample(snvs_valid, int(goodfrac * len(snvs_valid)))
self.indels = random.sample(indels_valid,
int(goodfrac * len(indels_valid)))
self.snvs += random.sample(
snvs_invalid, min(int(badfrac * len(snvs_valid)),
len(snvs_invalid)))
self.indels += random.sample(
indels_invalid,
min(int(badfrac * len(indels_valid)), len(indels_invalid)))
self.snvs = sorted(self.snvs)
self.indels = sorted(self.indels)
def build_sim(Ps, ms, es, incs, Mstar, Rstar):
assert Nplanets==len(ms)==len(es)==len(Ps)==len(incs)
Ws=2 * np.pi * rd.sample(Nplanets)
ws=2 * np.pi * rd.sample(Nplanets)
Ms=2 * np.pi * rd.sample(Nplanets)
radii = np.zeros(Nplanets)
for i in range(Nplanets):
radii[i] = np.cbrt(Ps[i] * Ps[i] * ms[i] / earth_mass_p_solar_mass / Mstar / 3) / 2
#set up simulation
sim = rebound.Simulation()
#add star
sim.add(m=Mstar, r=Rstar)
for i in range(Nplanets):
sim.add(m=ms[i] / earth_mass_p_solar_mass, P=Ps[i] / year_p_reboundtime, e=es[i], inc=incs[i], Omega=Ws[i], omega=ws[i], M=Ms[i], r=radii[i]) #G=1 units!
sim.move_to_com()
sim.collision = 'line'
sim.collision_resolve = collision
sim.ri_whfast.keep_unsynchronized = 1
sim.ri_whfast.safe_mode = 0
sim.integrator = "whfast"
sim.dt = np.sqrt(2) / 40 * sim.particles[1].P # ~0.035355
return sim
def update(self, u_t0, u_t1, x_t0):
"""
param[in] u_t0 : particle state odometry reading [x, y, theta] at time (t-1) [odometry_frame]
param[in] u_t1 : particle state odometry reading [x, y, theta] at time t [odometry_frame]
param[in] x_t0 : particle state belief [x, y, theta] at time (t-1) [world_frame]
param[out] x : particle state belief [x, y, theta] at time t [world_frame]
"""
ux_t0 = u_t0[0]
uy_t0 = u_t0[1]
uth_t0 = u_t0[2]
ux_t1 = u_t1[0]
uy_t1 = u_t1[1]
uth_t1 = u_t1[2]
del_rot1 = atan2(uy_t1-uy_t0,ux_t1-ux_t0) - uth_t0
del_trans = math.sqrt((ux_t1-ux_t0)^2 + (uy_t1-uy_t0)^2)
del_rot2 = uth_t1 - uth_t0 - del_rot1
del_rot1 = del_rot1 - sample(0,self.a1*del_rot1^2 + self.a2*del_trans^2)
del_trans = del_trans - sample(0, self.a3*del_trans^2 _ self.a4*(del_rot1^2 + del_rot2^2))
del_rot2 = del_rot2 - sample(0, self.a1*del_rot2^2 + self.a2*del_trans^2)
x_t1 = x_t0
x_t1[0] = x_t0[0] + del_trans*math.cos(x_t0[2] + del_rot1)
x_t1[1] = x_t0[1] + del_trans*math.cos(x_t0[2] + del_rot1)
x_t1[2] = x_t0[2] + del_rot1 +del_rot2
return x_t1
def create_single_odorant_activation_matrix(self):
"""
create_single_odorant_activation_matrix
"""
activation_matrix = np.zeros(
(self.params['n_patterns'], self.params['n_or']))
p1, p2, p3 = self.set_odorant_distribution_params()
np.random.seed(self.params['seed_activation_matrix'])
random.seed(self.params['seed_activation_matrix'])
distances = np.zeros((self.params['n_patterns'], self.params['n_or']))
n_min_active_OR = int(
round(self.params['n_or'] * self.params['frac_min_active_OR']))
n_max_active_OR = int(
round(self.params['n_or'] * self.params['frac_max_active_OR']))
n_above_thresh = np.zeros(
self.params['n_patterns']
) # number of glomeruli expected to get an activation larger than thresh
thresh = 0.01
for pn in xrange(self.params['n_patterns']):
n_active_OR = np.random.randint(n_min_active_OR, n_max_active_OR)
activated_ORs = random.sample(range(self.params['n_or']),
n_active_OR)
while np.unique(activated_ORs).size != n_active_OR:
activated_ORs = random.sample(range(self.params['n_or']),
n_active_OR)
for OR in activated_ORs:
dist = self.odorant_odor_distance_distribution((p1, p2, p3))
distances[pn, OR] = dist
affinity = np.exp(
-(dist)**2 / self.
params['distance_affinity_transformation_parameter_exp'])
if affinity > 1.:
affinity = 1.
if affinity < 0.:
affinity = 0.
activation_matrix[pn, OR] = affinity
n_above_thresh[pn] = (activation_matrix[pn, :] >
thresh).nonzero()[0].size
print 'ActivationMatrix: per pattern number of activated glom: mean %.2f +- %.2f, \t%.2f +- %.2f percent' % (n_above_thresh.mean(), n_above_thresh.std(), \
(n_above_thresh.mean() / float(self.params['n_or'])) * 100., (n_above_thresh.std() / float(self.params['n_or'])) * 100.)
if self.params['OR_activation_normalization']:
for pn in xrange(self.params['n_patterns']):
# this normalization increases the likelihood of having ORs with a high affinity to
# each given pattern --> receptors have specialized to odorants (?)
# if activation_matrix[pn, :].sum() == 0:
# print '\n\tWARNING: activation_matrix.sum for pattern %d == 0' % (pn)
activation_matrix[pn, :] /= activation_matrix[pn, :].max()
if not activation_matrix[pn, :].sum() == 0:
activation_matrix[pn, :] /= activation_matrix[pn, :].sum()
# pass
# for OR in xrange(self.params['n_or']):
# activation_matrix[:, OR] /= activation_matrix[:, OR].sum()
print 'Activation matrix sum:', activation_matrix.sum()
print "Activation matrix fn:", self.params['activation_matrix_fn']
np.savetxt(self.params['activation_matrix_fn'], activation_matrix)
return activation_matrix
def part1():
#Load the MNIST digit data
M = loadmat("mnist_all.mat")
f, axarr = plt.subplots(10, 10)
for i in range(10):
train_num = "train" + str(i)
test_num = "test" + str(i)
train_size = M[train_num].shape[0]
test_size = M[test_num].shape[0]
print(train_size)
random.seed(i)
rand_train_img = random.sample(range(0, train_size), 8)
rand_test_img = random.sample(range(0, test_size), 2)
# print(i)
# print(rand_train_img)
# print(rand_test_img)
# Display 8 training images and 2 test images.
for k in range(len(rand_train_img)):
img = M[train_num][k].reshape((28,28))
axarr[i, k].imshow(img, cmap=cm.gray)
axarr[i, k].axis('off')
for k in range(len(rand_test_img)):
img = M[train_num][k + 8].reshape((28,28))
axarr[i, k + 8].imshow(img, cmap=cm.gray)
axarr[i, k + 8].axis('off')
plt.show()
def k_plus(self, ace_prob):
if self.prenode is not None and self.prenode.status == "m" and sample() < self.K_PLUS:
# preNode is methylated and sample() gets smaller than K_PLUS
return MHistoneWithDNAModel(inherited=True, inherited_hst=self)
if self.nextnode is not None and self.nextnode.status == "m" and sample() < self.K_PLUS:
# nextNode is methylated and sample() gets smaller than KPLUS
return MHistoneWithDNAModel(inherited=True, inherited_hst=self)
return self
def __uniformMutation(self, population, npop):
"""
Operateur de mutation
"""
probaMutation = rd.sample((npop, self.__ndof))
deltaX = self.__stepMut * (rd.sample((npop, self.__ndof)) - 0.5)
population = population + deltaX * (probaMutation <= self.__rateMut)
return population
def creating_session(subsession):
for player in subsession.get_players():
# randomize to treatments
if player.round_number == 1:
player.participant.treatment = random.choice(
['Control', 'Emotional'])
player.treatment = player.participant.treatment
# print('set player.treatment to', player.treatment)
if player.round_number > 1:
prev_player = player.in_round(player.round_number - 1)
if player.round_number > Constants.num_rounds / 2:
player.sReward = 'LR'
# print(os.getcwd())
memelist = os.listdir('_static/LR')[1:-1]
# ! if you do not put the [1:-1] it crashes because there is a
# ! HIDDEN file inside of the mac folder called 'DS store' that breaks it lmao :D
# print(memelist)
pattern = r"meme(?P<number>\d{3})\.jpg"
# you are telling the pattern of the files inside of the folder, in my case a string...
# ...that says meme, then a THREE DIGIT number + .jpg, and then you group it by the number
# m = re.match(pattern, "meme200.jpg")
# print(m)
# int(m.group("number"))
numbers = [
int(re.match(pattern, x).group("number")) for x in memelist
]
# take all of the numbers from the image files and put them on a list
vImages = random.sample(numbers, 6)
# select 6 random numbers from the list, but they do not repeat each other
player.iImgPost1 = vImages[0]
player.iImgPost2 = vImages[1]
player.iImgPost3 = vImages[2]
player.iImgPost4 = vImages[3]
player.iImgPost5 = vImages[4]
player.iImgPost6 = vImages[5]
# player.iImgPost6 = random.randint(low=101,high=len(os.listdir('_static/LR')))
# depending on reward we check different images...
else:
player.sReward = 'HR'
# dRange = range(1,len(os.listdir('_static/HR')))
# vImages = random.sample(range(dRange), 6)
vImages = random.sample(range(1, len(os.listdir('_static/HR'))), 6)
#! make a subsample of 6 random images that are not the same and store the images in a vector
player.iImgPost1 = vImages[0]
player.iImgPost2 = vImages[1]
player.iImgPost3 = vImages[2]
player.iImgPost4 = vImages[3]
player.iImgPost5 = vImages[4]
player.iImgPost6 = vImages[5]
#save image post in the variable as one of the positions of the vector
player.iImgFeed = random.randint(1, 89)
def sample_dehnen(N, M, a, core=False):
# The factor M * 200^2 / 201^2 restricts the radius to 200 * a.
radii = dehnen_inverse_cumulative(
nprand.sample(N) * ((M * 40000) / 40401), M, a, core)
thetas = np.arccos(nprand.sample(N) * 2 - 1)
phis = 2 * pi * nprand.sample(N)
xs = radii * sin(thetas) * cos(phis)
ys = radii * sin(thetas) * sin(phis)
zs = radii * cos(thetas)
coords = np.column_stack((xs, ys, zs))
return coords
def set_bulge_positions():
factor = 0.9*M_bulge
radii = dehnen_inverse_cumulative(nprand.sample(N_bulge) * factor,
M_bulge, a_bulge, bulge_core)
thetas = np.arccos(nprand.sample(N_bulge)*2 - 1)
phis = 2 * pi * nprand.sample(N_bulge)
xs = radii * sin(thetas) * cos(phis)
ys = radii * sin(thetas) * sin(phis)
zs = radii * cos(thetas)
coords = np.column_stack((xs, ys, zs))
return coords
def set_bulge_positions():
factor = 0.9*M_bulge
radii = dehnen_inverse_cumulative(nprand.sample(N_bulge) * factor,
M_bulge, a_bulge, bulge_core)
thetas = np.arccos(nprand.sample(N_bulge)*2 - 1)
phis = 2 * pi * nprand.sample(N_bulge)
xs = radii * sin(thetas) * cos(phis)
ys = radii * sin(thetas) * sin(phis)
zs = radii * cos(thetas)
coords = np.column_stack((xs, ys, zs))
return coords
def sample_dehnen(N, M, a, core=False):
# The factor M * 200^2 / 201^2 restricts the radius to 200 * a.
radii = dehnen_inverse_cumulative(nprand.sample(N) *
((M*40000) / 40401), M, a, core)
thetas = np.arccos(nprand.sample(N)*2 - 1)
phis = 2 * pi * nprand.sample(N)
xs = radii * sin(thetas) * cos(phis)
ys = radii * sin(thetas) * sin(phis)
zs = radii * cos(thetas)
coords = np.column_stack((xs, ys, zs))
return coords
def next_genome_with_dna_model(hst_list, window, nuc_prob, ace_prob, p_off):
"""
this method takes histone list and returns the next generation of them.
"""
ahst_n = 0
mhst_n = 0
for i, hst in enumerate(hst_list):
if -window // 2 <= hst.position <= window // 2:
if hst.status == "a":
ahst_n += 1
elif hst.status == "m":
mhst_n += 1
for index in range(len(hst.CpGislandlist)):
# p_on probability
if hst.status == 'm' and sample() < 0.001133:
hst.CpGislandlist[index] = 1
# p_off probability
if sample() < p_off:
hst.CpGislandlist[index] = 0
hst = HistoneWithDNAModel.k(hst, ace_prob)
hst_list[i] = hst
p_bool = False
if mhst_n > 2:
p_bool = True
t_bool = (ahst_n > 5) and (not p_bool)
"""
WINDOW is size 10(11 histones note that there is E0 between E(-1) and E(1)),
so acetylated histones will be dominant if non-acetylated histones are less than 5.
"""
# if transcription does not happend, then with k_nuc
# probability, we recover E0 to be methylated.
eext_bool = False
if t_bool is False and sample() < nuc_prob:
eext_bool = True
# if in the locus we have more than two methylated histones,
# then with 100% prob, we recover E0 histone to be
# methylated. this is a histone memory part.
if mhst_n > 2:
eext_bool = True
if eext_bool:
center = len(hst_list) // 2
hst_list[center] = MHistoneWithDNAModel(inherited=True, inherited_hst=hst_list[center])
return hst_list, t_bool, p_bool
def set_disk_positions(N, z0):
radii = np.zeros(N)
# The maximum radius is restricted to 20 kpc.
sample = nprand.sample(N) * disk_radial_cumulative(20)
for i, s in enumerate(sample):
radii[i] = disk_radial_inverse_cumulative(s)
zs = disk_height_inverse_cumulative(nprand.sample(N), z0)
phis = 2 * pi * nprand.sample(N)
xs = radii * cos(phis)
ys = radii * sin(phis)
coords = np.column_stack((xs, ys, zs))
return coords
def set_disk_positions(N, z0):
radii = np.zeros(N)
# The maximum radius is restricted to 60 kpc.
sample = nprand.sample(N) * disk_radial_cumulative(60)
for i, s in enumerate(sample):
radii[i] = disk_radial_inverse_cumulative(s)
zs = disk_height_inverse_cumulative(nprand.sample(N), z0)
phis = 2 * pi * nprand.sample(N)
xs = radii * cos(phis)
ys = radii * sin(phis)
coords = np.column_stack((xs, ys, zs))
return coords
def __init__(self, genesis, network, nodeid, target=None):
self.genesis = genesis
self.network = network
self.peers = []
self.latencies = []
self.nodeid = nodeid
# A salt for this node, so all nodes don't produce the same hashes
self.nodesalt = uint256(sha256(randint(2**63 - 1)))
self.nonce = 0 # Will be increased in the mining process
self.reset(target)
# Geospatial location information
self.latitude = pi * (1 / 2 - sample(1))
self.longitude = 2 * pi * sample(1)
def generate_samples(self, model):
'''
Generate (s,a,r,s') pairs according to uniform distribution
'''
states = model.S()
result = []
for i in range(self._n_samples):
s = random.sample(states, 1)[0]
a = random.sample(model.A(s), 1)[0]
r = model.R(s, a)
s_p = util.functions.sample(model.T(s, a))
result.append((s, a, r, s_p))
return result
def generate_samples(self, model):
'''
Generate (s,a,r,s') pairs according to uniform distribution
'''
states = model.S()
result = []
for i in range(self._n_samples):
s = random.sample(states,1)[0]
a = random.sample(model.A(s),1)[0]
r = model.R(s,a)
s_p = util.functions.sample( model.T(s,a) )
result.append( (s,a,r,s_p) )
return result
def train_recommender(y, lamb, n):
m_users = y.shape[0]
m_champs = y.shape[1]
init_flat_theta = random.sample((m_users, n)).flatten() / 1000
init_flat_x = random.sample((m_champs, n)).flatten() / 1000
init_theta_and_x = np.concatenate((init_flat_x, init_flat_theta))
costfn = gen_costfn(y, lamb, n, m_users, m_champs)
theta_and_x = minimize(costfn, init_theta_and_x, method='BFGS',
options={'xtol': 1e-8, 'disp': True})
theta, x = unpack(theta_and_x)
return (theta, x)
def create_activation_matrix_for_pattern_completion_training(
self, n_active_OR):
"""
Create random patterns with a fixed number of activated OR per pattern
--> for pattern completion :
n_active_OR = int(round(params['n_or'] * params['frac_max_active_OR']))
--> for rivalry:
n_active_OR = int(round(params['n_or'] * params['frac_min_active_OR']))
"""
activation_matrix = np.zeros(
(self.params['n_patterns'], self.params['n_or']))
p1, p2, p3 = self.set_odorant_distribution_params()
np.random.seed(self.params['seed_activation_matrix'])
random.seed(self.params['seed_activation_matrix'])
distances = np.zeros((self.params['n_patterns'], self.params['n_or']))
for pn in xrange(self.params['n_patterns']):
activated_ORs = random.sample(range(self.params['n_or']),
n_active_OR)
while np.unique(activated_ORs).size != n_active_OR:
activated_ORs = random.sample(range(self.params['n_or']),
n_active_OR)
for OR in activated_ORs:
# draw a random distance from the fitted distance distribution
dist = self.odorant_odor_distance_distribution((p1, p2, p3))
distances[pn, OR] = dist
affinity = np.exp(
-(dist)**2 / self.
params['distance_affinity_transformation_parameter_exp'])
if affinity > 1.:
affinity = 1.
if affinity < 0.:
affinity = 0.
activation_matrix[pn, OR] = affinity
if self.params['OR_activation_normalization']:
for pn in xrange(self.params['n_patterns']):
# this normalization increases the likelihood of having ORs with a high affinity to
# each given pattern --> receptors have specialized to odorants (?)
activation_matrix[pn, :] /= activation_matrix[pn, :].max()
if not activation_matrix[pn, :].sum() == 0:
activation_matrix[pn, :] /= activation_matrix[pn, :].sum()
else:
print '\n\tWARNING: activation_matrix.sum for pattern %d == 0\nWill now quit, because that makes no sense' % (
pn)
exit(1)
print 'Activation matrix sum:', activation_matrix.sum()
print "Activation matrix fn:", self.params['activation_matrix_fn']
np.savetxt(self.params['activation_matrix_fn'], activation_matrix)
return activation_matrix
def generate_lon_lat(size):
"""Generate arrays of longitude, latitude.
Parameters
----------
size : int
length of arrays to generate
Returns
-------
list of 2 arrays
"""
return [random.sample(size) * 360 - 180, random.sample(size) * 180 - 90]
def __polynomialMutation(self, population, npop):
"""
Operateur de mutation
"""
probaMutation = rd.sample((npop, self.__ndof))
uMut = rd.sample((npop, self.__ndof))
uinf_filter = uMut <= 0.5
deltaMut = np.zeros_like(uMut)
deltaMut[uinf_filter] = (2 * uMut[uinf_filter])**self.__alpham - 1
deltaMut[~uinf_filter] = 1 - (2 *
(1 - uMut[~uinf_filter]))**self.__alpham
population = population + deltaMut * (probaMutation <= self.__rateMut)
return population
def graphs(self):
n = randint(100, 500)
# n = 4
print(n)
a = sample((n, n))
b = sample(n)
a * 100
b * 100
a = list(a)
b = list(b)
self.carousel.add_widget(plot_time_threads(n, a, b))
self.carousel.add_widget(plot_time_eps(n, a, b))
self.carousel.add_widget(plot_time_msize())
def set_bulge_positions():
global bulge_cut_M
bulge_cut_M = dehnen_cumulative(bulge_cut_r, M_bulge, a_bulge, gamma_bulge)
print "%.0f%% of bulge mass cut by the truncation..." % \
(100*(1-bulge_cut_M/M_bulge))
if bulge_cut_M < 0.9*M_bulge:
print " \_ Warning: this is more than 10%% of the total bulge mass!"
radii = dehnen_inverse_cumulative(nprand.sample(N_bulge) * bulge_cut_M,
M_bulge, a_bulge, gamma_bulge)
thetas = np.arccos(nprand.sample(N_bulge)*2 - 1)
phis = 2 * pi * nprand.sample(N_bulge)
xs = radii * sin(thetas) * cos(phis)
ys = radii * sin(thetas) * sin(phis)
zs = radii * cos(thetas)
coords = np.column_stack((xs, ys, zs))
return coords
def createRandomHistoneList(percentage=50,A=1,
NUM_OF_HISTONE=81,BEFORE_PROMOTER=40,
K_PLUS=0.12,
K_MINUS=0.117,
K_PLUS2=0.12,
K_ACE=0.12):
"""
percentage ... the probability of having methylated hitone.
"""
dstList = []
ratio = percentage/100 ## ratio should be float number between 0 and 1
for i in range(NUM_OF_HISTONE):
if(sample() < ratio):dstList.append(MHistone(position=i-BEFORE_PROMOTER,
K_PLUS=K_PLUS,
K_PLUS2=K_PLUS2,
K_MINUS=K_MINUS,
K_ACE=K_ACE,
A_bool=A,percentage=percentage))
else:dstList.append(AHistone(position=i-BEFORE_PROMOTER,
K_PLUS=K_PLUS,
K_PLUS2=K_PLUS2,
K_MINUS=K_MINUS,
K_ACE=K_ACE,
A_bool=A,percentage=percentage))
# dstList[i-1].display()
dstList[i-1].set_adjHistone(dstList[i])
dstList[NUM_OF_HISTONE-1].set_adjHistone(dstList[0]) ## Connect the head to tail
return dstList
def successor(self, method='reverse'):
if method == 'reverse':
ind = sorted(random.sample([i for i, _ in enumerate(self.path)],
2))
new_path = self.path[:]
new_path = new_path[:ind[0]] + new_path[
ind[0]:ind[1]][::-1] + new_path[ind[1]:]
return TravelingSalesmanProblem(new_path)
elif method == 'permutation':
new_path = self.path[:]
random.shuffle(new_path)
return TravelingSalesmanProblem(new_path)
elif method == 'adjacent':
successors = []
for i in range(len(self.path) - 1):
new_problem = self.copy()
new_problem.path[i], new_problem.path[
i + 1] = new_problem.path[i + 1], new_problem.path[i]
successors.append(new_problem)
last_path = self.copy()
last_path.path[0], last_path.path[-1] = last_path.path[
-1], last_path.path[0]
successors.append(last_path)
return random.choice(successors)
else:
print('No valid method supplied')
return False
def ramd():
sample_size = 500
rn1 = npr.rand(sample_size, 3)
rn2 = npr.randint(0, 10, sample_size)
rn3 = npr.sample(size=sample_size)
a = [0, 25, 50, 75, 100]
rn4 = npr.choice(a, size=sample_size)
fig, ((ax1, ax2), (ax3, ax4)) = plt.subplots(nrows=2,
ncols=2,
figsize=(10, 8))
ax1.hist(rn1, bins=25, stacked=True)
ax1.set_title('rand')
ax1.set_ylabel('frequency')
ax2.hist(rn2, bins=25)
ax2.set_title('randint')
ax3.hist(rn3, bins=25)
ax3.set_title('samople')
ax3.set_ylabel("frequency")
ax4.hist(rn4, bins=25)
ax4.set_title('choice')
plt.show()
def __init__(self, filename=None, winname=None, debug=False):
self.filename = filename
self.debug = debug
if not winname:
self.winname = "{:%Y%m%d_%H%M%S}_{}".format(
datetime.now(), sample())
else:
self.winname = winname
if filename:
print("[Load param]", filename)
with open(filename) as fp:
config = json.load(fp)
self.config = config
if debug:
cv.namedWindow(self.winname)
if "default" in self.config:
for trackbarname, [min_val, val, max_val
] in sorted(config["default"].items()):
self._create_default(trackbarname, min_val, val, max_val)
if "kernel" in self.config:
for trackbarname, [min_val, val, max_val
] in sorted(config["kernel"].items()):
self._cerate_kernel(trackbarname, min_val, val, max_val)
self._create_save("save", 0, 0, 1)
def sample_relation(self, poi, bounding_box, perspective, landmark, step=0.02):
"""
Sample a relation given a point of interest and landmark.
Evaluate each relation and probabilisticaly choose the one that is likely to
generate the poi given a landmark.
"""
rel_scores = []
rel_classes = []
rels = []
for s in [DistanceRelationSet, OrientationRelationSet, ContainmentRelationSet]:
for rel in s.relations:
rel_scores.append(self.evaluate_poi(poi, bounding_box, rel, perspective, landmark, step))
rel_classes.append(rel)
rel_scores = array(rel_scores)
set_printoptions(threshold='nan')
# print 'X',rel_scores
rel_probabilities = rel_scores/sum(rel_scores)
# print 'X',rel_probabilities
index = rel_probabilities.cumsum().searchsorted( random.sample(1) )[0]
# print 'X',index
# print 'X',rel_probabilities.cumsum()
return rel_classes[index], rel_probabilities[index], self.get_entropy(rel_probabilities)
def fit(
self,
trial: np.ndarray,
show: bool = False,
guess: np.ndarray = None,
) -> 'CoinMixture':
"""fits a coin mixture model via EM
:param trial: m_flips each trial
:type trial: np.ndarray
:param show: to show the results of iteration,
defaults to False
:type show: bool, optional
:return: fitted object
:rtype: CoinMixture
"""
# intial guess for probabilities associated with each coin
if guess is None:
guess = sample(self.n_coins)
theta = [0] * self.max_iter
# iterate
for i in range(self.max_iter):
theta[i] = chain(guess)
if show:
print(f"#{i} ", ", ".join(f"{c:.4f}" for c in chain(guess)))
# compute the e-step
num_heads, num_tails = self.estep(trial, guess)
# compute the m-step
guess = self.mstep(num_heads, num_tails)
self.theta = pd.DataFrame(theta)
return self
def transform(self, ratio=None):
shape = list(self.shape)
if ratio is None:
ratio = self.ratio
# if n_rows is None:
# ratio = 0.5
# elif n_rows < 1:
# ratio = n_rows
# else:
# ratio = 1.0*n_rows/shape[0]
shape[0] *= ratio
ud_X = random.sample(shape) * self.gap_row + self.min_row
X = self.X
Y = np.zeros(len(X), dtype=np.int)
ud_Y = np.ones(len(ud_X), dtype=np.int)
data_X = np.concatenate((X, ud_X), axis=0)
target_Y = np.concatenate((Y, ud_Y), axis=0)
return data_X, target_Y
def initialize(n,power):
# Returns a Python dictionary representing a web
# with n pages, and where each page k is linked to by
# L_k random other pages. The L_k are independent and
# identically distributed random variables with a
# shifted and truncated Pareto probability mass function
# p(l) proportional to 1/(l+1)^power.
# The representation used is a Python dictionary with
# keys 0 through n-1 representing the different pages.
# i[j][0] is the estimated PageRank, initially set at 1/n,
# i[j][1] the number of outlinks, and i[j][2] a list of
# the outlinks.
# This dictionary is used to supply (key,value) pairs to
# both mapper tasks defined below.
# initialize the dictionary
i = {}
for j in xrange(n): i[j] = [1.0/n,0,[]]
# For each page, generate inlinks according to the Pareto
# distribution. Note that this is somewhat tedious, because
# the Pareto distribution governs inlinks, NOT outlinks,
# which is what our representation is adapted to represent.
# A smarter representation would give easy
# access to both, while remaining memory efficient.
for k in xrange(n):
lk = paretosample(n+1,power)-1
values = random.sample(xrange(n),lk)
for j in values:
i[j][1] += 1 # increment the outlink count for page j
i[j][2].append(k) # insert the link from j to k
return i
def init_genome(percentage=50,
hst_n=81,
kp=0.176,
km=0.117):
"""
percentage ... the probability of having methylated hitone.
this method returns a list of histone randomly generated with respect to
the inputs.
"""
before_promoter = hst_n // 2
hst_list = [] # hst_list stores histones
ratio = percentage / 100 # ratio should be float number between 0 and 1
for i in range(hst_n):
if sample() < ratio:
hst_list.append(MHistone(position=i - before_promoter,
kp=kp,
km=km,
)
)
else:
hst_list.append(UHistone(position=i - before_promoter,
kp=kp,
km=km,
)
)
hst_list[i - 1].set_adjhistone(hst_list[i])
hst_list[0].prenode = None # disjoint the edge histone to itself.
return hst_list
def predict_node(node, inner_bins, root_bins, leaf_bins, is_root=False):
"""
Predict label for current node and continue recursively
"""
correct = total = 0
if node.is_leaf:
node.fprop = True
left = right = None
else:
if not node.left.fprop:
corr, tot, left = Baseline.predict_node(node.left, inner_bins, root_bins, leaf_bins)
correct += corr
total += tot
if not node.right.fprop:
corr, tot, right = Baseline.predict_node(node.right, inner_bins, root_bins, leaf_bins)
correct += corr
total += tot
node.fprop = True
bins = root_bins if is_root else leaf_bins if node.is_leaf else inner_bins
label = np.digitize(sample(1), bins)[0]
pred = Node(label)
pred.word = node.word
if node.is_leaf:
pred.is_leaf = True
else:
pred.left = left
pred.right = right
left.parent = pred
right.parent = pred
return correct + (label == node.label), total + 1, pred
def pg_polygons_wgs84():
size = 1000
x1, y1 = generate_lon_lat(size)
x2, y2 = generate_lon_lat(size)
# generate some fields in the data frame
f = random.sample(size) * 360 - 180
i = random.randint(-32767, 32767, size=size)
ui = random.randint(0, 65535, size=size).astype("uint64")
df = DataFrame(
data={
"x1": x1,
"y1": y1,
"x2": x2,
"y2": y2,
"f": f,
"i": i,
"ui": ui,
"labels": i.astype("str"),
})
# Generate random triangles
df["geometry"] = df[["x1", "y1", "x2", "y2"]].apply(
lambda row: pg.polygons([[row.x1, row.y1], [row.x2, row.y1],
[row.x2, row.y2], [row.x1, row.y1]]),
axis=1,
)
return df
def init_genome_with_dna_model(percentage=50,
hst_n=81,
kp=0.176,
km=0.117):
"""
this method is a modified version of createRandomHistoneList
used for Oct4 histones.
"""
before_promoter = hst_n // 2
hst_list = [] # hst_list stores histones
ratio = percentage / 100 # ratio should be float number between 0 and 1
for i in range(hst_n):
if sample() < ratio:
hst_list.append(MHistoneWithDNAModel(position=i - before_promoter,
kp=kp,
km=km,
)
)
else:
hst_list.append(UHistoneWithDNAModel(position=i - before_promoter,
kp=kp,
km=km,
)
)
hst_list[i - 1].set_adjhistone(hst_list[i])
hst_list[0].prenode = None # disjoint the edge histone to itself.
return hst_list
def get_stock_list_two_strategies(strategy,amount):
#define the stocks for each strategy
stocks = stock_info[strategy]
stock_list_multiple=[]
# basic_data={}
random_stocks= random.sample(stocks,2)
# basic_data["strategy"]=strategy
# stock_list_multiple.append(basic_data)
for stock in random_stocks:
stock_list={}
print(stock)
response_data= requests.get('https://api.iextrading.com/1.0/stock/'+stock+'/quote')
response_week = requests.get('https://www.alphavantage.co/query?function=TIME_SERIES_DAILY&symbol='+stock+'&outputsize=compact&apikey=6Z2XPX3WLFSGNNPI')
response_data2=response_data.json()
response_week2=response_week.json()
print(response_data2['companyName'])
stock_list["symbolName"]=response_data2['symbol']
stock_list["companyName"]=response_data2['companyName']
stock_list["latestPrice"]=response_data2['latestPrice']
stock_list["changePercentage"] =response_data2['changePercent']
stock_list["investmentAmount"]=str(int(amount)/4)
# stock_list["weeklyData"]=response_week2["Time Series (Daily)"]
stock_list["weeklyData"]=response_week2["Time Series (Daily)"]
stock_list["strategy"]=strategy
stock_list_multiple.append(stock_list)
# the stocks_list for the selected strategies
return stock_list_multiple
def initWords(phonemes,conceptCount,wordLengths):
newWords = []
for _ in range(conceptCount):
newWords.append(frozenset(random.sample(phonemes, wordLengths))) # must be frozen (hashable) so networkx can make nodes out of words
#for _ in range(wordCount):
# newWords.append([random.choice(phonemes) for _ in range(wordLengths)])
return newWords
def set_halo_positions():
global halo_cut_M
halo_cut_M = dehnen_cumulative(halo_cut_r, M_halo, a_halo, gamma_halo)
print "%.0f%% of halo mass cut by the truncation..." % \
(100*(1-halo_cut_M/M_halo))
if halo_cut_M < 0.9*M_halo:
print " \_ Warning: this is more than 10%% of the total halo mass!"
radii = dehnen_inverse_cumulative(nprand.sample(N_halo) * halo_cut_M,
M_halo, a_halo, gamma_halo)
thetas = np.arccos(nprand.sample(N_halo)*2 - 1)
phis = 2 * pi * nprand.sample(N_halo)
xs = radii * sin(thetas) * cos(phis)
ys = radii * sin(thetas) * sin(phis)
zs = radii * cos(thetas)
coords = np.column_stack((xs, ys, zs))
return coords
def __init__(self, name, snvs_valid, snvs_invalid, indels_valid, indels_invalid, sensitivity=None, precision=None):
self.name = name
self.sensitivity = random.uniform(0.5, 0.95) if not sensitivity else sensitivity
self.precision = random.uniform(0.5, 0.95) if not precision else precision
goodfrac = self.sensitivity
badfrac = self.sensitivity*(1.-self.precision)/self.precision
self.snvs = random.sample(snvs_valid, int(goodfrac*len(snvs_valid)))
self.indels = random.sample(indels_valid, int(goodfrac*len(indels_valid)))
self.snvs += random.sample(snvs_invalid, min(int(badfrac*len(snvs_valid)), len(snvs_invalid)))
self.indels += random.sample(indels_invalid, min(int(badfrac*len(indels_valid)), len(indels_invalid)))
self.snvs = sorted(self.snvs)
self.indels = sorted(self.indels)
def pd_create_dataset(seed=None):
""" make sample data """
if seed == None:
seed = random.randint(low=0, high=1000)
random.seed(seed)
pool1 = ['boy', 'girl']
pool2 = ['blue', 'red', 'green', 'yellow', 'purple']
pool3 = pd.date_range('1/1/2014', periods=1000)
d = {
'num': [random.randint(low=0, high=1000) for i in range(50)],
'num2': [
random.randint(low=0, high=1000) + random.sample()
for i in range(50)
],
'str':
[pool1[random.randint(low=0, high=len(pool1))] for i in range(50)],
'str2':
[pool2[random.randint(low=0, high=len(pool2))] for i in range(50)],
'date':
[pool3[random.randint(low=0, high=len(pool3))] for i in range(50)]
}
df = pd.DataFrame(d)
return seed, df
def weightfunc():
i_weight = np.zeros((10,784))
for i in range(0,10):
x = (1-(-1))*random.sample(784) - 1
i_weight[i] = x
weight_matrix=np.matrix(np.array(i_weight))
return weight_matrix
def normalMutation(population, delta_mut, rho_mut):
npop, ndof = population.shape
probaMutation = rd.sample((npop, ndof))
deltaX = delta_mut * (rd.normal(size=(npop, ndof)))
population = population + deltaX * (probaMutation <= rho_mut)
return population
def __SBXcrossover(self, selection, population, npop):
couples = np.zeros((npop // 2, 2, self.__ndof))
children = np.zeros_like(population)
uCross = rd.sample((npop // 2, self.__ndof))
betaCross = np.zeros_like(uCross)
uinf_filter = uCross <= 0.5
betaCross[uinf_filter] = (2 * uCross[uinf_filter])**(self.__alphac)
betaCross[~uinf_filter] = (2 * (1 - uCross[~uinf_filter]))**(
-self.__alphac)
for i in range(npop // 2):
k = i
while k == i:
k = rd.randint(0, npop // 2 - 1)
couples[i] = [selection[i], selection[k]]
x1 = couples[:, 0]
x2 = couples[:, 1]
children[:npop // 2] = 0.5 * ((1 - betaCross) * x1 +
(1 + betaCross) * x2)
children[npop // 2:] = 0.5 * ((1 + betaCross) * x1 +
(1 - betaCross) * x2)
return children
def forward(self, src, trg, teacher_forcing_ratio=0.5):
# src = [src sent len, batch size]
# trg = [trg sent len, batch size]
# teacher_forcing_ratio is probability to use teacher forcing
# e.g. if teacher_forcing_ratio is 0.75 we use ground-truth inputs 75% of the time
# Again, now batch is the first dimension instead of zero
max_len, batch_size = trg.shape
trg_vocab_size = self.decoder.output_dim
# tensor to store decoder outputs
outputs = torch.zeros(max_len,
batch_size,
trg_vocab_size,
device=self.device)
# last hidden state of the encoder is used as the initial hidden state of the decoder
enc_states, hidden, cell = self.encoder(src)
if USE_BIDIR_ENCODER:
hidden = hidden.mean(0).unsqueeze(0)
cell = cell.mean(0).unsqueeze(0)
# first input to the decoder is the <sos> tokens
input = trg[0]
use_teacher_forcing = rnd.sample(max_len - 1) < teacher_forcing_ratio
for t, teacher_forcing in enumerate(use_teacher_forcing, 1):
#print(enc_states.shape, hidden.shape, cell.shape)
output, hidden, cell = self.decoder(input, hidden, cell,
enc_states)
outputs[t] = output
input = (trg[t] if teacher_forcing else output.max(1)[1])
return outputs
def sample_landmark(class_, landmarks, poi):
distances = array([lmk.distance_to(poi) for lmk in landmarks])
scores = 1.0/(array(distances)**1.5 + class_.epsilon)
scores[distances == 0] = 0
lm_probabilities = scores/sum(scores)
index = lm_probabilities.cumsum().searchsorted( random.sample(1) )[0]
return index
def sample_point_trajector(self, bounding_box, relation, perspective, landmark, step=0.02):
"""
Sample a point of interest given a relation and landmark.
"""
probs, points = self.get_probabilities_box(bounding_box, relation, perspective, landmark)
probs /= probs.sum()
index = probs.cumsum().searchsorted( random.sample(1) )[0]
return Landmark( 'point', Vec2( *points[index] ), None, Landmark.POINT )
def set_disk_positions(N, z0):
global disk_cut
radii = np.zeros(N)
disk_cut = disk_radial_cumulative(disk_cut_r)
print "%.0f%% of disk mass cut by the truncation..." % \
(100*(1-disk_cut))
if disk_cut < 0.9:
print " \_ Warning: this is more than 10% of the total disk mass!"
sample = nprand.sample(N) * disk_cut
for i, s in enumerate(sample):
radii[i] = disk_radial_inverse_cumulative(s)
zs = disk_height_inverse_cumulative(nprand.sample(N), z0)
phis = 2 * pi * nprand.sample(N)
xs = radii * cos(phis)
ys = radii * sin(phis)
coords = np.column_stack((xs, ys, zs))
return coords
def project_size():
seed = random.sample()
if seed < 0.1:
return 0
elif seed < 0.85:
return 1
else:
return 2
def sample_relation(self, sampled_landmark, poi):
rel_scores = []
for relation in self.relations:
rel_scores.append( relation.probability(poi, sampled_landmark) )
rel_scores = array(rel_scores)
rel_probabilities = rel_scores/sum(rel_scores)
index = rel_probabilities.cumsum().searchsorted( random.sample(1) )[0]
return self.relations[index]
def variants_from_genome(fasta, n_snvs, n_indels, max_indel_size):
"""
Reads a genome file (.fasta), selects n valid snvs and indels
given the parameters.
"""
valid_chromosomes = [str(i) for i in range(1, 22)] + ["X", "Y"]
genome = readfasta(fasta)
genome = [(chrom, seq) for chrom, seq in genome \
if chrom in valid_chromosomes]
sizes = numpy.array([len(seq) for chrom, seq in genome])
cum_sizes = numpy.cumsum(sizes)
snv_starts = sorted(random.sample(range(cum_sizes[-1]), n_snvs))
indel_starts = sorted(random.sample(range(cum_sizes[-1]), n_indels))
return snvs_from_starts(genome, snv_starts, cum_sizes), \
indels_from_starts(genome, indel_starts, cum_sizes, max_indel_size)
def k_minus(self):
"""
the methylated histone will be get unmethylated by K_MINUS probability
or
the acetilated histone will be get unacetylated by K_MINUS probability
return unmethylated histone object if the histone will get unmethylated
"""
if(sample()<Histone.K_MINUS):return UHistone(copy=True,copy_histone=self)
return self
def sample_poi(self, bounding_box, relation, perspective, landmark, step=0.02):
"""
Sample a point of interest given a relation and landmark.
"""
#points = landmark.representation.sample_points(step=step)
#probs = self.get_probabilities_points(points, relation, perspective, landmark)
probs, points = self.get_probabilities_box(bounding_box, relation, perspective, landmark)
probs /= probs.sum()
index = probs.cumsum().searchsorted( random.sample(1) )[0]
return points.flatten()[index]
def weighted_sample(iterable, weights, n=None):
if len(iterable) != len(weights):
raise Exception('Length mismatch')
count = len(iterable) if n is None else n
# Normalize weights
wt_sum = float(sum(weights))
aliases = array(w / wt_sum for w in weights).cumsum()
# Get indices to include in sample
indices = aliases.searchsorted(sample(count))
return array(iterable[i] for i in indices)
def sample_landmark(self, landmarks, poi):
''' Weight by inverse of distance to landmark center and choose probabilistically '''
epsilon = 0.000001
distances = array([lmk.representation.middle.distance_to(poi) for lmk in landmarks])
scores = 1.0/(array(distances)**1.5 + epsilon)
# scores[distances == 0] = 0
lm_probabilities = scores/sum(scores)
index = lm_probabilities.cumsum().searchsorted( random.sample(1) )[0]
return landmarks[index], lm_probabilities[index], self.get_entropy(lm_probabilities)
def k_ace(self, ace_prob):
# if histone has CpG island on it
if 1 in self.CpGislandlist:
# histone cannnot get acetylated
return self
else:
if sample() < ace_prob:
return AHistoneWithDNAModel(inherited=True, inherited_hst=self)
else:
return self
