def nc_of_expon():
E1 = lambda x : x*0.5*np.exp(-x/2)
E2 = lambda x : x**2*0.5*np.exp(-x/2)
print integrate.quad(E1, 0, np.inf)
print integrate.quad(E2, 0, np.inf)
print expon(scale=2).moment(1)
print expon(scale=2).var()
def metro_exp_poison(chute=[1], N=1000):
valores = chute
taxa = []
priori = expon(1)
for i in range(N):
expo_aux = expon(1)
valor = expo_aux.rvs()
U = random.rand()
x_dado_y = poisson(valores[-1])
y_dado_x = poisson(valor)
teste = ( priori.pdf(valor) * x_dado_y.pmf(int(valores[-1])) ) / ( priori.pdf(valores[-1]) * y_dado_x.pmf(int(valor)) )
if min([teste,1]) > U:
valores.append(valor)
taxa.append(1)
else:
valores.append(valores[-1])
taxa.append(0)
return {"valores":valores , "taxa":sum(taxa)/len(taxa)}
def testExponentialSampleMultiDimensional(self):
with self.test_session():
batch_size = 2
lam_v = [3.0, 22.0]
lam = constant_op.constant([lam_v] * batch_size)
exponential = exponential_lib.Exponential(rate=lam)
n = 100000
samples = exponential.sample(n, seed=138)
self.assertEqual(samples.get_shape(), (n, batch_size, 2))
sample_values = samples.eval()
self.assertFalse(np.any(sample_values < 0.0))
for i in range(2):
self.assertLess(
stats.kstest(
sample_values[:, 0, i],
stats.expon(scale=1.0 / lam_v[i]).cdf)[0],
0.01)
self.assertLess(
stats.kstest(
sample_values[:, 1, i],
stats.expon(scale=1.0 / lam_v[i]).cdf)[0],
0.01)
def do_train_rand(train, valid, params=None, max_models=32):
"""Do randomized hyper-parameter search
Args:
train (SFrame): training set
valid (SFrame): validataion set
params (dict): parameters for random search
max_models (int): maximum number of models to run
Returns:
res (SFrame): table of choices of parameters sorted by valid RMSE
"""
if not params:
params = {'user_id': ['username'], 'item_id': ['course_id'],
'target': ['label'], 'binary_target': [True],
'num_factors': stats.randint(4, 128),
'regularization': stats.expon(scale=1e-4),
'linear_regularization': stats.expon(scale=1e-7)}
try:
job = gl.toolkits.model_parameter_search \
.random_search.create((train, valid),
gl.recommender.
factorization_recommender.create,
params, max_models=max_models)
res = job.get_results()
res = res.sort('validation_rmse')
print 'Best params for random search are: {}'.format(res[0])
res.save('rand_search.csv', format='csv')
except:
print job.get_metrics()
res = None
return res
def get_param_grid(cur_model, points, rand):
print('\nRetrieving parameter grid...')
try:
c_range = 10.0 ** np.arange(-2, 3)
# print 'Getting Parameter grid...'
# out_txt.write('Getting Parameter grid...')
gamma_range = [0, .01, .1, .3]
# neighbor_range = np.arange(2, points, step=5)
# leaf_range = np.arange(10, points, step=5)
neighbor_range = np.arange(2, 17, step=5)
leaf_range = np.arange(10, 60, step=5)
if not rand:
grid_params = {'SVC()': [{'C': c_range,
'kernel': ['poly'],
'degree': [3, 5, 8],
'gamma': gamma_range,
'probability': [True],
'class_weight': ['auto', None]},
{'C': c_range,
'kernel': ['rbf', 'sigmoid'],
'gamma': gamma_range,
'probability': [True],
'class_weight': ['auto', None]},
{'C': c_range,
'kernel': ['linear'],
'random_state': [10],
'probability': [True],
'class_weight': ['auto', None]}],
'KNeighborsClassifier()': [{'n_neighbors': neighbor_range,
'weights': ['uniform'],
'algorithm': ['brute'],
'metric': ['euclidean', 'manhattan']},
{'n_neighbors': neighbor_range,
'weights': ['uniform'],
'algorithm': ['ball_tree', 'kd_tree'],
'metric': ['euclidean', 'manhattan'],
'leaf_size': leaf_range}],
'LogisticRegression()': [{'penalty': ['l1', 'l2'],
'C': c_range,
'class_weight': [None, 'auto']}]}
return grid_params[cur_model]
else:
rand_params = {'SVC()': {'C': stats.expon(scale=300),
'kernel': ['linear', 'poly', 'rbf', 'sigmoid'],
'degree': [3, 4, 5, 6, 7, 8],
'gamma': stats.expon(scale=1/3),
'random_state': [10],
'probability': [True],
'class_weight': ['auto', None]},
'KNeighborsClassifier()': {'n_neighbors': stats.randint(low=2, high=20),
'weights': ['uniform', 'distance'],
'algorithm': ['ball_tree', 'kd_tree', 'brute'],
'metric': ['euclidean', 'manhattan'],
'leaf_size': stats.randint(low=10, high=60)},
'LogisticRegression()': {'penalty': ['l1', 'l2'],
'C': stats.expon(scale=300),
'class_weight': [None, 'auto']}}
return rand_params[cur_model]
except:
print('could not get parameter grid')
def generate_receiver_events(self, trial_start_time, trial_duration_seconds, num_receivers_at_time_zero, arrival_rate, service_rate):
"""Generates receiver init and termination events.
Receiver initialization events are generated as a poission process with arrival rate: arrival_rate (in receivers per second).
Each receiver has an exponential service time with rate: service_rate.
This should be called at the start of a simulation run, just after initialization of mininet.
"""
if self.event_list is not None:
return
self.event_list = []
self.past_event_list = []
self.trial_start_time = trial_start_time
# First, generate the initial receievers (active at time 0)
for i in range(0, num_receivers_at_time_zero):
# Generate a service time
service_time = expon(loc = 0, scale=(1.0 / service_rate)).rvs(1)[0]
# Select a host through a uniform random distribution
receiver = self.net_hosts[randint(0,len(self.net_hosts))]
receiver = MulticastReceiverApplication(receiver, self.group_ip, self.mcast_port, self.echo_port, trial_start_time, service_time)
self.receiver_applications.append(receiver)
self.event_list.append((trial_start_time, DynamicMulticastGroupDefinition.EVENT_RECEIVER_INIT, receiver))
self.event_list.append((trial_start_time + service_time, DynamicMulticastGroupDefinition.EVENT_RECEIVER_TERMINATION, receiver))
# Alternative: Generate inter-arrival times using exponential distribution
arrival_times = []
expo_rv = expon(loc = 0, scale=(1.0 / arrival_rate))
last_arrival_time = 0
while last_arrival_time < trial_duration_seconds:
next_arrival_time = expo_rv.rvs(1)[0] + last_arrival_time
if next_arrival_time < trial_duration_seconds:
arrival_times.append(next_arrival_time)
last_arrival_time = next_arrival_time
# Alternative Method
# Find the number of arrivals in the interval [0, trial_duration_seconds]
# Size = trial_duration_seconds, since we want the number of arrivals in trial_duration_seconds time units
# num_arrivals = sum(poisson.rvs(arrival_rate, size=trial_duration_seconds))
# Once the number of arrivals is known, generate arrival times uniform on [0, trial_duration_seconds]
# arrival_times = []
# for i in range(0, num_arrivals):
# arrival_times.append(uniform(0, trial_duration_seconds))
# Now, for each arrival, generate a corresponding receiver application and events
for arrival_time in arrival_times:
# Generate a service time
service_time = expon(loc = 0, scale=(1.0 / service_rate)).rvs(1)[0]
# Select a host through a uniform random distribution
receiver = self.net_hosts[randint(0,len(self.net_hosts))]
receiver = MulticastReceiverApplication(receiver, self.group_ip, self.mcast_port, self.echo_port, trial_start_time + arrival_time, service_time)
self.receiver_applications.append(receiver)
self.event_list.append((trial_start_time + arrival_time, DynamicMulticastGroupDefinition.EVENT_RECEIVER_INIT, receiver))
self.event_list.append((trial_start_time + arrival_time + service_time, DynamicMulticastGroupDefinition.EVENT_RECEIVER_TERMINATION, receiver))
# Sort the event list by time
self.event_list = sorted(self.event_list, key=lambda tup: tup[0])
# Debug printing
for event in self.event_list:
print 'Time:' + str(event[0]) + ' ' + str(event[1]) + ' ' + str(event[2])
def test_randomized_search_grid_scores():
# Make a dataset with a lot of noise to get various kind of prediction
# errors across CV folds and parameter settings
X, y = make_classification(n_samples=200, n_features=100, n_informative=3, random_state=0)
# XXX: as of today (scipy 0.12) it's not possible to set the random seed
# of scipy.stats distributions: the assertions in this test should thus
# not depend on the randomization
params = dict(C=expon(scale=10), gamma=expon(scale=0.1))
n_cv_iter = 3
n_search_iter = 30
search = RandomizedSearchCV(SVC(), n_iter=n_search_iter, cv=n_cv_iter, param_distributions=params, iid=False)
search.fit(X, y)
assert_equal(len(search.grid_scores_), n_search_iter)
# Check consistency of the structure of each cv_score item
for cv_score in search.grid_scores_:
assert_equal(len(cv_score.cv_validation_scores), n_cv_iter)
# Because we set iid to False, the mean_validation score is the
# mean of the fold mean scores instead of the aggregate sample-wise
# mean score
assert_almost_equal(np.mean(cv_score.cv_validation_scores), cv_score.mean_validation_score)
assert_equal(list(sorted(cv_score.parameters.keys())), list(sorted(params.keys())))
# Check the consistency with the best_score_ and best_params_ attributes
sorted_grid_scores = list(sorted(search.grid_scores_, key=lambda x: x.mean_validation_score))
best_score = sorted_grid_scores[-1].mean_validation_score
assert_equal(search.best_score_, best_score)
tied_best_params = [s.parameters for s in sorted_grid_scores if s.mean_validation_score == best_score]
assert_true(
search.best_params_ in tied_best_params,
"best_params_={0} is not part of the" " tied best models: {1}".format(search.best_params_, tied_best_params),
)
def nc_of_expon():
# 1st non-center moment of expon distribution whose lambda is 0.5
E1 = lambda x: x * 0.5 * np.exp(-x / 2)
# 2nd non-center moment of expon distribution whose lambda is 0.5
E2 = lambda x: x ** 2 * 0.5 * np.exp(-x / 2)
print(integrate.quad(E1, 0, np.inf))
print(integrate.quad(E2, 0, np.inf))
print(expon(scale=2).moment(1))
print(expon(scale=2).moment(2))
def SalehValenzuela(**kwargs):
""" generic Saleh and Valenzuela Model
Parameters
----------
Lam : clusters Poisson Process parameter (ns)
lam : rays Poisson Process parameter (ns)
Gam : clusters exponential decay factor
gam : rays exponential decay factor
T : observation duration
"""
defaults = { 'Lam' : .1,
'lam' : .5,
'Gam' : 30,
'gam' : 5 ,
'T' : 100}
for k in defaults:
if k not in kwargs:
kwargs[k]=defaults[k]
Lam = kwargs['Lam']
lam = kwargs['lam']
Gam = kwargs['Gam']
gam = kwargs['gam']
T = kwargs['T']
Nr = 1.2*T/Lam
Nc = 1.2*T/lam
e1 = st.expon(1./Lam)
e2 = st.expon(1./lam)
# cluster time of arrival
tc = np.cumsum(e1.rvs(Nr))
tc = tc[np.where(tc<T)]
Nc = len(tc)
tauc = np.kron(tc,np.ones((1,Nr)))[0,:]
# rays time of arrival
taur = np.cumsum(e2.rvs((Nr,Nc)),axis=0).ravel()
# exponential decays of cluster and rays
etc = np.exp(-tauc/(1.0*Gam))
etr = np.exp(-taur/(1.0*gam))
et = etc*etr
tau = tauc+taur
# filtering < T and reordering in delay domain
tau = tau[np.where(tau<T)]
et = et[np.where(tau<T)]
u = np.argsort(tau)
taus = tau[u]
ets = et[u]*np.sign(np.random.rand(len(u))-0.5)
SVir = bs.TBsignal(taus,ets)
return(SVir)
def testExponential1(self):
distV = expon(scale=5)
distW = expon(scale=3)
v = 2
w = 3
expected = ([0,1],[0,3])
bids = [v,w]
distributions = [distV,distW]
obtained = myersonAuction(bids,distributions)
self.assertAlmostEqual(expected, obtained,
msg="Myerson auction with inputs: " + str(bids) + ", " +
str(distributions) + ". Expected " + str(expected) +
"but obtained " + str(obtained) + ".")
def sgd(pd, pl, qd, ql):
params = {'loss':['squared_loss', 'huber', 'epsilon_insensitive',
'squared_epsilon_insensitive'],
'alpha':expon(scale=1),
'epsilon':expon(scale=1),
'l1_ratio':uniform(),
'penalty':[ 'l2', 'l1', 'elasticnet']}
clf = SGDRegressor()
#clf = RandomizedSearchCV(clf, params, n_jobs=2, n_iter=10, verbose=10)
print("Training Linear SVM Randomly")
clf.fit(pd, pl)
print("Score: " + str(clf.score(qd, ql)))
return clf
def __init__(self, scenario_flag = "Freeway_Free"):
"""
Totally five scenarios are supported here:
Freeway_Night, Freeway_Free, Freeway_Rush;
Urban_Peak, Urban_Nonpeak.
The PDFs of the vehicle speed and the inter-vehicle space are adapted
from existing references.
"""
if scenario_flag == "Freeway_Night":
self.headway_random = expon(0.0, 1.0/256.41)
meanSpeed = 30.93 #m/s
stdSpeed = 1.2 #m/s
elif scenario_flag == "Freeway_Free":
self.headway_random = lognorm(0.75, 0.0, np.exp(3.4))
meanSpeed = 29.15 #m/s
stdSpeed = 1.5 #m/s
elif scenario_flag == "Freeway_Rush":
self.headway_random = lognorm(0.5, 0.0, np.exp(2.5))
meanSpeed = 10.73 #m/s
stdSpeed = 2.0 #m/s
elif scenario_flag == "Urban_Peak":
scale = 1.096
c = 0.314
loc = 0.0
self.headway_random = fisk(c, loc, scale)
meanSpeed = 6.083 #m/s
stdSpeed = 1.2 #m/s
elif scenario_flag == "Urban_Nonpeak":
self.headway_random = lognorm(0.618, 0.0, np.exp(0.685))
meanSpeed = 12.86 #m/s
stdSpeed = 1.5 #m/s
else:
raise
self.speed_random = norm(meanSpeed, stdSpeed)
def test_random_vector(self):
comp = (stats.expon(), stats.beta(0.4, 0.8), stats.norm())
rv = best.random.RandomVectorIndependent(comp)
print str(rv)
x = rv.rvs()
print 'One sample: ', x
print 'pdf:', rv.pdf(x)
x = rv.rvs(size=10)
print '10 samples: ', x
print 'pdf: ', rv.pdf(x)
print rv.mean()
print rv.var()
print rv.std()
print rv.stats()
# Split it in two:
rv1, rv2 = rv.split(0)
print str(rv1)
x = rv1.rvs(size=5)
print x
print rv1.pdf(x)
print rv2.pdf(x)
print str(rv2)
print x
x = rv2.rvs(size=5)
print rv2.pdf(x)
rv3, rv4 = rv1.split(0)
print str(rv3)
print str(rv4)
rv5, rv6 = rv3.split(1)
print str(rv5)
print str(rv6)
rv7, rv8 = rv5.split(2)
print str(rv7)
print str(rv8)
def gillespie_logistique2(taille_ini, b1,b2,d1,d2,temps):
"""une autre implémentation de l'algorithme de Gillepie
on ne conserve la taille qu'à des instants prédéfinis"""
taille = zeros(temps.size) # préalocation de la mémoire
# initialisation des temps et taille courantes
temps_courant, taille_courante = 0.0, taille_ini
t_nais = (b1 + b2 * taille_courante) * taille_courante # taux de naissance
t_mort = (d1 + d2 * taille_courante) * taille_courante # taux de mort
tau = t_nais + t_mort # taux global
ee = expon()
uu = uniform()
delta_temps = ee.rvs() / tau
for k in range(temps.size):
# on simule sans dépasser temps[k]
while temps_courant + delta_temps < temps[k]:
temps_courant += delta_temps # mise à jour instant courant
if uu.rvs() < (b1 * taille_courante) / tau:
taille_courante += 1 # naissance
else:
taille_courante -= 1 # mort
t_nais = (b1 + b2 * taille_courante) * taille_courante # taux de naissance
t_mort = (d1 + d2 * taille_courante) * taille_courante # taux de mort
tau = t_nais + t_mort # taux global
delta_temps = ee.rvs() / tau # temps de séjour
taille[k] = taille_courante
return taille
def test_conditional_rv(self):
return
px = stats.expon()
py = best.random.RandomVariableConditional(px, (1, 2),
name='Conditioned Exponential')
print str(py)
print py.rvs(size=10)
print py.interval(0.5)
print py.median()
print py.mean()
print py.var()
print py.std()
print py.stats()
print py.moment(10)
return
i = (0, 4)
t = np.linspace(i[0], i[1], 100)
plt.plot(t, py.pdf(t), t, py.cdf(t), linewidth=2.)
plt.legend(['PDF', 'CDF'])
#plt.show()
py1, py2 = py.split()
print str(py1)
print str(py2)
plt.plot(t, py1.pdf(t), t, py1.cdf(t),
t, py2.pdf(t), t, py2.cdf(t), linewidth=2)
plt.legend(['PDF $y_1$', 'CDF $y_1$', 'PDF $y_2$', 'CDF $y_2$'])
def get_value(self):
fig = figure()
xname = "invariantMass"
xmin,xmax,xbins = -5.,15.,50
index = "run*"
x,counts = vis_bokeh.get_1d_hist(xname,xmin,xmax,xbins,es,index=index)
deltas = np.sqrt(counts)
fig=vis_bokeh.whiskered_histogram(xmin,xmax,xbins,counts,deltas,-deltas)
#fit params
model = mix
model.fit(sample_weight=counts)#,values_init={'sig_weightlog':np.log(0.4),'bck_weightlog':np.log(0.6)})
parameters = model.parameters
w_sig,w_bkg =np.exp(parameters['sig_weightlog']),np.exp(parameters['bck_weightlog'])
w_sum = w_sig+w_bkg
w_sig,w_bkg = w_sig/w_sum,w_bkg/w_sum
n_events = np.sum(counts)
norm = n_events*(xmax-xmin)/xbins
#plot lines
expo = lambda x_arr:st.expon(0,1./parameters['slope']).pdf(x_arr)*w_bkg*norm
gauss = lambda x_arr:st.norm(parameters['mean'],parameters['sigma']).pdf(x_arr)*w_sig*norm
pdf_x = np.arange(1000,dtype='float')/1000.*(xmax-xmin) + xmin
fig.line(pdf_x, expo(pdf_x), legend="Background", line_width=2,color = 'red')
fig.line(pdf_x, gauss(pdf_x), legend="Signal", line_width=2,color='blue')
fig.line(pdf_x, gauss(pdf_x)+expo(pdf_x), legend="Sum", line_width=2,color='green')
fig.xaxis.axis_label = time.strftime("%H:%M:%S")
return vis_bokeh.fig_to_html(fig)
def fit_exponential(vect):
exponential_dist = stats.expon(scale=mean(vect)) # lambda = 1 / mean
print 'Performing the KS test on exponential data:'
dstat, p = stats.kstest(vect, exponential_dist.pdf)
print 'D-Statistic:\t{}'.format(dstat)
print 'P Value:\t\t{}'.format(p)
print '----------------------------'
def particlesF(t,pRecs,tol,xs,sim,distFunc,ii):
'''
# Filter particles step.
'''
p_num = len(pRecs)
np.random.seed()
sys.stdout.flush()
a_star = np.zeros(p_num)
rho = tol+1.
N = np.size(pRecs[0],axis=1) #number of particles. Should check all pRec match.
n = 1000 #upper bound for number of samples rejected before raising error.
rejects = 0 #count number of rejects.
while(rho>tol and rejects < n):
#FIXED: if no particles are accepted after a number of steps tolerance may be too low
# fix by adding condition to raise error after n particles being rejected.
r = np.random.randint(0,high=N)
for i in range(p_num):
if (pRecs[i][t-1,r] > 0):
a_star[i] = stats.gamma.rvs(pRecs[i][t-1,r]/rw_var,scale=rw_var)#p1A[t-1,r] + stats.norm.rvs(scale=0.1)
else:
a_star[i] = stats.expon(scale=0.1).rvs()
ys = sim(*a_star)#sim(a_star[0],a_star[1],ii)
rho = distFunc(ys,xs)
rejects += 1
if (rejects >= n):
raise NameError('Rejected all particles. Try increasing tolerances or increasing number of particles to reject.')
res = a_star.tolist()
res.append(rho)
return res #return parameters and accepted distance.
def svc_appr():
"""
Best params: {'C': 0.022139881953014046}
Submission:
E_val:
E_in:
E_out:
"""
from sklearn.svm import LinearSVC
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
from sklearn.cross_validation import StratifiedKFold
from sklearn.grid_search import RandomizedSearchCV
from scipy.stats import expon
X, y = dataset.load_train()
raw_scaler = StandardScaler()
raw_scaler.fit(X)
X_scaled = raw_scaler.transform(X)
svc = LinearSVC(dual=False, class_weight='auto')
rs = RandomizedSearchCV(svc, n_iter=50, scoring='roc_auc', n_jobs=-1,
cv=StratifiedKFold(y, 5), verbose=2,
param_distributions={'C': expon()})
rs.fit(X_scaled, y)
logger.debug('Got best SVC.')
logger.debug('Best params: %s', rs.best_params_)
logger.debug('Grid scores:')
for i, grid_score in enumerate(rs.grid_scores_):
print('\t%s' % grid_score)
logger.debug('Best score (E_val): %s', rs.best_score_)
logger.debug('E_in: %f', Util.auc_score(rs, X_scaled, y))
def setup_logistic():
"""
Creates clf pipeline for Logistic Regression
Returns pipeline and parameters for GridSearchCV
"""
pipeline = Pipeline(steps=[('scaler', MinMaxScaler()),
('kbest', SelectKBest(score_func=f_classif)),
('pca', PCA()),
('clf', LogisticRegression())
]
)
params = {'kbest__k': range(3,30),
'pca__whiten': (True, False),
# 'clf__C': [ 0.001, 0.1, 10, 10**2, 10**5, 10**10],
'clf__C': expon(),
'clf__class_weight': [{False: 1, True: 12},
{False: 1, True: 10},
{False: 1, True: 8},
{False: 1, True: 6},
{False: 1, True: 4}],
'clf__tol': [2**i for i in range(-20, -1)],
'clf__penalty': ('l1','l2')
}
return pipeline, params
def create_svm(pd, pl, qd, ql):
lsvc = LinearSVC()
params = {'C': expon(scale=100)}
svm = RandomizedSearchCV(lsvc, params, n_jobs=4, n_iter=10, verbose=10)
print("Training Linear SVM Randomly")
svm.fit(pd, pl)
print("SVM Score: " + str(svm.score(qd, ql)))
return svm
def exponential(location = 0.0, scale = 1.0, N = None, quantiles = None):
# Exponential distribution
# similar usage to scipy.stats.expon(loc, scale)
# The scale parameter is equal to scale = 1.0 / lambda
# see http://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.expon.html
from scipy.stats import expon
ppf_engine = expon(location, scale).ppf
return continuous(ppf = ppf_engine, quantiles = quantiles, N = N, Str = 'exponential distribution with location = %g and scale = %g' % (location, scale))
def __init__(self, rate_func, dt):
SpikingInputStream.__init__(self)
self.dt = dt
self.length = 1
self.rate_func = rate_func
self.spikes = np.ones([1, SPIKE_HISTORY_LENGTH])*-10000.0
self.rv_expon = expon()
self.reset()
def test_random_vector_independent(self):
return
comp = (stats.expon(), stats.beta(0.4, 0.8), stats.norm())
rv = best.random.RandomVectorIndependent(comp)
subdomain = [[0.1, 4.], [0.1, 0.8], [-1., 1.]]
rvc = best.random.RandomVectorConditional(rv, subdomain)
print str(rvc)
return
def nc_of_expon():
rv = expon(scale=2)
print(rv.mean())
print(rv.var())
print(rv.moment(1))
print(rv.moment(2))
print(rv.moment(3))
print(rv.moment(4))
print(rv.stats(moments='mvsk'))
def Exp(lamda, tag=None):
"""
An Exponential random variate
Parameters
----------
lamda : scalar
The inverse scale (as shown on Wikipedia), FYI: mu = 1/lamda.
"""
return uv(rv=ss.expon(scale=1./lamda), tag=tag)
def test_random_search_cv_results():
# Make a dataset with a lot of noise to get various kind of prediction
# errors across CV folds and parameter settings
X, y = make_classification(n_samples=200, n_features=100, n_informative=3,
random_state=0)
# scipy.stats dists now supports `seed` but we still support scipy 0.12
# which doesn't support the seed. Hence the assertions in the test for
# random_search alone should not depend on randomization.
n_splits = 3
n_search_iter = 30
params = dict(C=expon(scale=10), gamma=expon(scale=0.1))
random_search = RandomizedSearchCV(SVC(), n_iter=n_search_iter,
cv=n_splits, iid=False,
param_distributions=params)
random_search.fit(X, y)
random_search_iid = RandomizedSearchCV(SVC(), n_iter=n_search_iter,
cv=n_splits, iid=True,
param_distributions=params)
random_search_iid.fit(X, y)
param_keys = ('param_C', 'param_gamma')
score_keys = ('mean_test_score', 'mean_train_score',
'rank_test_score',
'split0_test_score', 'split1_test_score',
'split2_test_score',
'split0_train_score', 'split1_train_score',
'split2_train_score',
'std_test_score', 'std_train_score',
'mean_fit_time', 'std_fit_time',
'mean_score_time', 'std_score_time')
n_cand = n_search_iter
for search, iid in zip((random_search, random_search_iid), (False, True)):
assert_equal(iid, search.iid)
cv_results = search.cv_results_
# Check results structure
check_cv_results_array_types(cv_results, param_keys, score_keys)
check_cv_results_keys(cv_results, param_keys, score_keys, n_cand)
# For random_search, all the param array vals should be unmasked
assert_false(any(cv_results['param_C'].mask) or
any(cv_results['param_gamma'].mask))
check_cv_results_grid_scores_consistency(search)
def __init__(self, totaljobs, numServers, labda, mu, failurerate, maintenance):
np.random.seed(1)
self.rho = labda/mu
self.scheduler = Scheduler()
now = self.scheduler.now
self.sender = Sender()
arrival = expon(scale = 1./labda)
service = expon(scale = 1./mu)
self.sender.setTotalJobs( totaljobs )
self.sender.setTimeBetweenConsecutiveJobs(arrival)
self.queue = Server()
self.queue.setServiceTimeDistribution(service)
self.sink = Sink()
self.sender.Out = self.queue
self.queue.In = self.sender
self.queue.Out = self.sink
self.sink.In = self.queue
self.scheduler.register(self.sender, self.queue)
self.sender.start()
def get_simulation(dv=.001, update_method='approx', approx_order=None, tol=1e-8):
import scipy.stats as sps
# Create simulation:
b1 = ExternalPopulation(100)
i1 = InternalPopulation(v_min=0, v_max=.02, dv=dv, update_method=update_method, approx_order=approx_order, tol=tol)
b1_i1 = Connection(b1, i1, 1, delays=0.0, weights=(sps.expon(0,.005), 201))
simulation = Network([b1, i1], [b1_i1])
return simulation
def test_random_vector_poly(self):
return
comp = (stats.expon(), stats.beta(0.4, 0.8), stats.norm())
rv = best.random.RandomVectorIndependent(comp)
print str(rv)
prod = best.gpc.ProductBasis(degree=5, rv=rv)
print str(prod)
x = rv.rvs(num_samples=10)
print x
print prod(x)
def ppf(self,p):
if self.rtype=="n":
return(norm(loc=self.args[0],scale=self.args[1]).ppf(p))
elif self.rtype=="ln":
return(lognorm(s=self.args[1],scale=math.exp(self.args[0])).ppf(p))
elif self.rtype=="g":
return(gumbel_r(loc=self.args[0],scale=self.args[1]).ppf(p))
elif self.rtype=="e":
return(expon(loc=self.args[0],scale=self.args[1]).ppf(p))
elif self.rtype=="u":
return(uniform_r(loc=self.args[0],scale=self.args[1]).ppf(p))
else:
print("distribution {0} not found" .format(rtype))
return("error - distribution")
def part2():
def create_exponential_estimator(k):
k_factorial = scipy.special.gamma(k)
def exponential_estimator(samples):
return (k_factorial / (samples**k).mean(axis=1))**1 / k
return exponential_estimator
exponential_estimators = [(create_exponential_estimator(k),
'$k = {}$'.format(k)) for k in range(1, 5 + 1)]
grid_for_tetta = np.arange(0.01, 5 + 0.01, 0.01)
draw_risk(grid_for_tetta, exponential_estimators,
lambda tetta: sps.expon(scale=1 / tetta))
def get_area_distribution(tracks, fit=False):
area = np.sum(tracks > 0, axis=(1, 2))
if not fit:
count = np.bincount(area)
probability = count / float(np.sum(count))
return stats.rv_discrete(a=0,
b=np.max(probability.shape[0]),
name='signal distribution',
values=(np.arange(count.shape[0]),
probability))
else:
exp_params = stats.expon.fit(area)
return stats.expon(*exp_params)
def shop_4():
N, L = 10, 100
G = uniform(loc=4, scale=2) # G is called a frozen distribution.
a = superposition(G.rvs((N, L)))
labda = 1.0 / 5
E = expon(scale=1.0 / (N * labda))
print(E.mean())
x, y = cdf(a)
dist_name = "U[4,6]"
plot_distributions(x, y, N, L, E, dist_name)
print(KS(a, E))
def histexponencial(self):
np.random.seed(2016) # replicar random
# parametros esteticos de seaborn
sns.set_palette("deep", desat=.6)
sns.set_context(rc={"figure.figsize": (8, 4)})
a = float(self.para.text())
exponencial = stats.expon(a)
aleatorios = exponencial.rvs(1000) # genera aleatorios
cuenta, cajas, ignorar = plt.hist(aleatorios, 20)
plt.ylabel('frequencia')
plt.xlabel('valores')
plt.title('Histograma Exponencial')
plt.show()
def KSTestCDFPlot(Stat_Stn, Stn, Setting):
MonthlyStat = Stat_Stn["MonthlyStat"]
Prep = Stat_Stn["PrepDF"]["P"]
fig, axs = plt.subplots(nrows=4, ncols=3, sharex=True, sharey=True)
m = 0
for i in range(3):
for j in range(4):
Prep_m = Prep[Prep.index.month == (m + 1)].dropna()
Prep_m = Prep_m[Prep_m != 0]
coef1 = MonthlyStat.loc[m + 1, "exp"]
coef2 = MonthlyStat.loc[m + 1, "gamma"]
coef3 = MonthlyStat.loc[m + 1, "weibull"]
coef4 = MonthlyStat.loc[m + 1, "lognorm"]
# Plot
ecdf = ECDF(Prep_m)
x = np.arange(0, max(Prep_m), 0.1)
axs[j, i].plot(x,
expon(coef1[0], coef1[1]).cdf(x),
label='exp',
linestyle=':')
axs[j, i].plot(x,
gamma(coef2[0], coef2[1], coef2[2]).cdf(x),
label='gamma',
linestyle=':')
axs[j, i].plot(x,
weibull_min(coef3[0], coef3[1], coef3[2]).cdf(x),
label='weibull',
linestyle=':')
xlog = np.arange(min(np.log(Prep_m)), max(np.log(Prep_m)), 0.1)
axs[j, i].plot(np.exp(xlog),
norm(coef4[0], coef4[1]).cdf(xlog),
label='lognorm',
linestyle=':')
axs[j, i].plot(ecdf.x, ecdf.y, label='ecdf', color="red")
axs[j,
i].axvline(x=130, color="black", linestyle="--",
linewidth=1) # Definition of storm defined by CWB
axs[j, i].set_title(str(m + 1))
axs[j, i].legend()
m += 1
fig.suptitle("KStest CDF " + Stn, fontsize=16)
# Add common axis label
fig.text(0.5, 0.04, 'Precipitation (mm)', ha='center', fontsize=14)
fig.text(0.05, 0.5, 'CDF', va='center', rotation='vertical', fontsize=14)
fig.set_size_inches(18.5, 10.5)
plt.tight_layout(rect=[0.06, 0.05, 0.94,
0.94]) #rect : tuple (left, bottom, right, top)
SaveFig(fig, "KStest CDF " + Stn, Setting)
plt.show()
return None
def _impose_white_noise(self, data):
import scipy.stats as stats
original_shape = data.shape
noise_area_distr = stats.expon(scale = self._white_noise_rate)
data = data.reshape(data.shape[0], -1)
s = data.shape[1]
for i in xrange(data.shape[0]):
n_white_noise = int(np.minimum(noise_area_distr.rvs(size=1), self._white_noise_maximum) * s)
indx = np.random.choice(s, size=n_white_noise, replace=False)
data[i, indx] = self._signal_level
return data.reshape(original_shape)
def simulation_4():
N = 1 # number of customers
L = 300
labda = 1.0 / 5 # lambda is a function in python. Hence we write labda
E = expon(scale=1.0 / labda)
print(E.mean()) # to check that we chose the right scale
a = E.rvs(L)
print(KS(a, E))
x, y = cdf(a)
dist_name = "U[4,6]"
plot_distributions(x, y, N, L, E, dist_name)
def shop_3():
N, L = 3, 100
G = uniform(loc=4, scale=2)
a = superposition(G.rvs((N, L)))
labda = 1.0 / 5
E = expon(scale=1.0 / (N * labda))
print(E.mean())
x, y = cdf(a)
dist_name = "U[4,6]"
plot_distributions(x, y, N, L, E, dist_name)
print(KS(a, E)) # Compute KS statistic using the function defined earlier
def testExponentialSampleMultiDimensional(self):
batch_size = 2
lam_v = [3.0, 22.0]
lam = tf.constant([lam_v] * batch_size)
exponential = tfd.Exponential(rate=lam, validate_args=True)
n = 100000
samples = exponential.sample(n, seed=test_util.test_seed())
self.assertEqual(samples.shape, (n, batch_size, 2))
sample_values = self.evaluate(samples)
self.assertFalse(np.any(sample_values < 0.0))
for i in range(2):
self.assertLess(
sp_stats.kstest(sample_values[:, 0, i],
sp_stats.expon(scale=1.0 / lam_v[i]).cdf)[0],
0.01)
self.assertLess(
sp_stats.kstest(sample_values[:, 1, i],
sp_stats.expon(scale=1.0 / lam_v[i]).cdf)[0],
0.01)
def __init__(self, pose, agent=None, sensor=None, color="black", \
noise_per_meter=5, noise_std=math.pi/60,\
bias_rate_stds=(0.1,0.1),\
expected_stuck_time=1e100, expected_escape_time=1e-100,\
expected_kidnap_time=1e100, kidnap_range_x=(-5.0,5.0), kidnap_range_y=(-5.0,5.0)):
super().__init__(pose,agent,sensor,color)
self.noise_pdf = expon(scale=1.0/(1e-100+noise_per_meter))#exponは指数分布の関数なので、パラメーターを入れるだけでよい(λを掛けなくてよい)
self.distance_until_noise = self.noise_pdf.rvs()
self.theta_noise = norm(scale=noise_std)
self.bias_rate_nu = norm.rvs(loc=1.0,scale=bias_rate_stds[0])
self.bias_rate_omega = norm.rvs(loc=1.0,scale=bias_rate_stds[1])
self.stuck_pdf = expon(scale=expected_stuck_time)#回数の逆数は間隔なのでそのまま入れる
self.escape_pdf = expon(scale=expected_escape_time)
self.time_until_stuck = self.stuck_pdf.rvs()
self.time_until_escape = self.escape_pdf.rvs()
self.is_stuck = False
self.kidnap_pdf = expon(scale=expected_kidnap_time)
self.time_until_kidnap = self.kidnap_pdf.rvs()
rx, ry =kidnap_range_x, kidnap_range_y
self.kidnap_dist = uniform(loc=(rx[0],ry[0],0.0),scale=(rx[1]-rx[0],ry[1]-ry[0],2*math.pi))#x,y,theta の3次元
def testExponentialSample(self):
lam = tf.constant([3.0, 4.0])
lam_v = [3.0, 4.0]
n = tf.constant(100000)
exponential = exponential_lib.Exponential(rate=lam)
samples = exponential.sample(n, seed=tfp_test_util.test_seed())
sample_values = self.evaluate(samples)
self.assertEqual(sample_values.shape, (100000, 2))
self.assertFalse(np.any(sample_values < 0.0))
for i in range(2):
self.assertLess(
sp_stats.kstest(sample_values[:, i],
sp_stats.expon(scale=1.0 / lam_v[i]).cdf)[0], 0.01)
def record_factory():
"""Generator for fake sales records."""
pid = 0
while True:
pid += 1
gender = random.choice(["male", "female"])
if gender == "male":
first_name = fkr.first_name_male()
else:
first_name = fkr.first_name_female()
data = {
"id":
pid,
"first_name":
first_name,
"last_name":
fkr.last_name(),
"birthdate":
fkr.date_between(start_date="-30y", end_date="-18y"),
"gender":
gender,
"city":
random.choice(
["Amsterdam", "Den Haag", "Eindhoven", "Utrecht",
"Rotterdam"]),
"orders":
1 + int(stats.expon(0.1).rvs() * 3),
"order_amount":
1 + int(stats.expon(0.1).rvs() * 20),
"opt_in":
random.choice([True, False]),
}
yield data
def __init__(self, pose, agent=None, sensor=None, color="black", \
noise_per_meter=5, noise_std=math.pi/60, \
bias_rate_stds=(0.1,0.1), \
expected_stuck_time = 1e100, expected_escape_time = 1e-100, \
expected_kidnap_time=1e100, kidnap_range_x=(-5.0,5.0), kidnap_range_y=(-5.0,5.0)):
super().__init__(pose, agent, sensor,
color) # IdeealRobotの__init__メソッドを呼び出す
# 踏み石用確率密度関数の作成（指数分布）
self.noise_pdf = expon(scale=1.0 / (1e-100 + noise_per_meter))
# 最初に小石を踏むまでの道のり
self.distance_until_noise = self.noise_pdf.rvs()
# thetaに加えるノイズ
self.theta_noise = norm(scale=noise_std)
# ロボット固有のバイアスの作成
self.bias_rate_nu = norm.rvs(loc=1.0, scale=bias_rate_stds[0])
self.bias_rate_omega = norm.rvs(loc=1.0, scale=bias_rate_stds[1])
# スタック用確率密度関数の作成（指数分布）
self.stuck_pdf = expon(scale=expected_stuck_time)
self.escape_pdf = expon(scale=expected_escape_time)
# 時間の初期化
self.time_until_stuck = self.stuck_pdf.rvs()
self.time_until_escape = self.escape_pdf.rvs()
# ロボットがスタック中が表すグラフ
self.is_stuck = False
# 誘拐が起こる確率密度関数（指数分布）
self.kidnap_pdf = expon(scale=expected_kidnap_time)
self.time_until_kidnap = self.kidnap_pdf.rvs()
# 誘拐後のロボット位置の範囲
rx, ry = kidnap_range_x, kidnap_range_y
# 誘拐後のロボットの位置・姿勢の確率密度関数（一様分布）
self.kidnap_dist = uniform(loc=(rx[0], ry[0], 0.0),
scale=(rx[1] - rx[0], ry[1] - ry[0],
2 * math.pi))
def random_point(self, shape):
"""
Sample uniformly from the constraint set.
L1 and L2 are implemented here.
Linf implemented in the subclass.
https://arxiv.org/abs/math/0503650
"""
if self.p == 2:
distrib = Normal(0, 1)
elif self.p == 1:
distrib = Laplace(0, 1)
x = distrib.sample(shape)
e = expon(.5).rvs()
denom = torch.sqrt(e + (x**2).sum())
return self.alpha * x / denom
def get_simulation(dv=.001, update_method='approx', tol=1e-8):
import scipy.stats as sps
# Create simulation:
b1 = ExternalPopulation(50)
b2 = ExternalPopulation(1000)
i1 = InternalPopulation(v_min=-.04,
v_max=.02,
dv=dv,
update_method=update_method,
tol=tol)
b1_i1 = Connection(b1,
i1,
1,
delays=0.0,
weights=(sps.expon(0, .00196), 301))
b2_i1 = Connection(b2,
i1,
1,
delays=0.0,
weights=(sps.expon(0, .001), 301))
simulation = Network([b1, b2, i1], [b1_i1, b2_i1])
return simulation
def checkFitService(vData, dMu, sDistribution):
dMax = np.max(vData)
vK = np.arange(dMax)
iScale = 1 / dMu
vDistribution = st.expon.pdf(vK, scale=iScale)
plt.figure()
plt.subplot(1, 2, 1)
plotDistribution(vData, vDistribution, sDistribution)
plt.subplot(1, 2, 2)
st.probplot(vData, dist=st.expon(scale=iScale), plot=plt)
plt.grid()
plt.tight_layout()
plt.show()
def generate_exp():
a = request.args.get('a', 0, type=float)
# generate pdf and return json results
thetas = np.linspace(0, 5, 200)
# expon is a standardized version of exponential dist
# set scale to 1/lambda for non-scaled version
prior = st.expon(scale= (1/a))
ydat = prior.pdf(thetas)
ycdf = prior.cdf(thetas)
validIndex = ~np.isinf(ydat)
d = collections.OrderedDict()
d['x'] = list(thetas[validIndex])
d['y'] = list(ydat[validIndex])
d['cdf'] = list(ycdf[validIndex])
return jsonify(result = d)
def IMPORTANCE_SAMPLING(q, Ns):
lbd1 = [lbdi + qi for qi, lbdi in zip(q, lbd)]
X = [expon(scale=1 / lbdi).rvs(Ns) for lbdi in lbd1]
W = [0] * Ns
C = [0] * Ns
for ns in range(Ns):
if ns != 0 and ns % 10**4 == 0: print(ns)
x = [Xi[ns] for Xi in X]
W[ns] = exp(CALC_LOGW(x, lbd, lbd1, T))
C[ns] = CALC_F(x)
# print ns, W[ns], C[ns], x
return W, C
def validlhs_regular():
from pyDOE import lhs as lhsd
n_samples = 0
while n_samples != 300:
lhs = lhsd(6, 2800)
#expand x, y coordinates to their real values
lhs[:, 0] = expon(scale=10).ppf(lhs[:, 0])
lhs[:, 1] = norm(0, 2.5).ppf(lhs[:, 1])
lhs[:, 2] = expon(scale=10).ppf(lhs[:, 2])
lhs[:, 3] = norm(0, 2.5).ppf(lhs[:, 3])
lhs[:, 4] = expon(scale=10).ppf(lhs[:, 4])
lhs[:, 5] = norm(0, 2.5).ppf(lhs[:, 5])
#exclude points where turbine 2 is closer than turbine 1
valid_1_2 = calculate_distance(lhs[:, 2],
lhs[:, 3]) \
- calculate_distance(lhs[:, 0], lhs[:, 1])
lhs = lhs[valid_1_2 > 0]
#exclude points where turbine 3 is closer than turbine 2
valid_2_3 = calculate_distance(lhs[:, 4],
lhs[:, 5]) \
- calculate_distance(lhs[:, 2], lhs[:, 3])
lhs = lhs[valid_2_3 > 0]
#exclude points where turbines are closer than 2D to origin
dist_1 = np.sqrt(lhs[:, 0]**2 + lhs[:, 1]**2)
lhs = lhs[dist_1 > 2]
dist_2 = np.sqrt(lhs[:, 2]**2 + lhs[:, 3]**2)
lhs = lhs[dist_2 > 2]
dist_3 = np.sqrt(lhs[:, 4]**2 + lhs[:, 5]**2)
lhs = lhs[dist_3 > 2]
#exclude points where turbines are closer than 2D from
#each other
dist_1_2 = np.sqrt((lhs[:, 0] - lhs[:, 2])**2 +
(lhs[:, 1] - lhs[:, 3])**2)
lhs = lhs[dist_1_2 > 2]
dist_2_3 = np.sqrt((lhs[:, 2] - lhs[:, 4])**2 +
(lhs[:, 3] - lhs[:, 5])**2)
lhs = lhs[dist_2_3 > 2]
dist_3_1 = np.sqrt((lhs[:, 4] - lhs[:, 0])**2 +
(lhs[:, 5] - lhs[:, 1])**2)
lhs = lhs[dist_3_1 > 2]
n_samples = len(lhs)
print(n_samples)
#return to transformed coordinates
lhs[:, 0] = expon(scale=10).cdf(lhs[:, 0])
lhs[:, 1] = norm(0, 2.5).cdf(lhs[:, 1])
lhs[:, 2] = expon(scale=10).cdf(lhs[:, 2])
lhs[:, 3] = norm(0, 2.5).cdf(lhs[:, 3])
lhs[:, 4] = expon(scale=10).cdf(lhs[:, 4])
lhs[:, 5] = norm(0, 2.5).cdf(lhs[:, 5])
# replace lhs points with nearest regular arrays
X_reg_tran = np.loadtxt('regular_arrays_no_rot_transformed.txt')
min_dist = np.zeros(len(X_reg_tran))
for i in range(len(lhs)):
diff = np.linalg.norm(X_reg_tran - lhs[i, :], axis=1)
lhs[i, :] = X_reg_tran[np.argmin(diff), :]
return lhs
def KRRModel(n_jobs=int(os.getenv("SLURM_CPUS_PER_TASK", 3)),
cv=5,
n_iter=30,
verbose=2,
best_params=None):
if best_params is None:
kr = RandomizedSearchCV(KernelRidge(kernel="rbf"),
param_distributions={
"alpha": expon(scale=.02),
"gamma": expon(scale=.06)
},
verbose=verbose,
n_jobs=n_jobs,
cv=cv,
n_iter=n_iter)
else:
if "kernel" not in best_params:
best_params["kernel"] = "rbf"
kr = KernelRidge(**best_params)
model = make_pipeline(StandardScaler(), kr)
return model
def __init__(self,
pose,
agent=None,
sensor=None,
color="black",
noise_per_meter=5,
noise_std=math.pi / 60):
super().__init__(pose, agent, sensor,
color) # IdeealRobotの__init__メソッドを呼び出す
# 指数分布のオブジェクト
self.noise_pdf = expon(scale=1.0 / (1e-100 + noise_per_meter))
# 最初に小石を踏むまでの道のり
self.distance_until_noise = self.noise_pdf.rvs()
# thetaに加えるノイズ
self.theta_noise = norm(scale=noise_std)
def __init__(self, numBeams=41, sparsity=1):
self.pHit = 0.95
self.pShort = 0.02
self.pMax = 0.02
self.pRand = 0.01
self.sigmaHit = 0.05
self.lambdaShort = 1
self.zMax = 20
self.zMaxEps = 0.02
self.Angles = np.linspace(-np.pi, np.pi, numBeams) # array of angles
self.Angles = self.Angles[::sparsity]
# Pre-compute for efficiency
self.normal = norm(0, self.sigmaHit)
self.exponential = expon(self.lambdaShort)
def mpdf(x, Ns, Nb, mu, sg, lb, comps=["sig", "bkg"]):
sig = norm(mu, sg)
sigN = np.diff(sig.cdf(mrange))
bkg = expon(mrange[0], lb)
bkgN = np.diff(bkg.cdf(mrange))
tot = 0
if "sig" in comps:
tot += Ns * sig.pdf(x) / sigN
if "bkg" in comps:
tot += Nb * bkg.pdf(x) / bkgN
return tot
def __init__(self, pose, agent=None, sensor=None, color="black",
noise_per_meter=5, noise_std=math.pi/60,
bias_rate_stds=(0.1, 0.1),
expected_stuck_time=1e100, expected_escape_time=1e-100,
expected_kidnap_time=1e100, kidnap_range_x=(-5.0, 5.0), kidnap_range_y=(-5.0, 5.0)):
super().__init__(pose, agent, sensor, color)
# noise #
self.noise_pdf = expon(scale=1.0/(1e-100 + noise_per_meter))
self.distance_until_noise = self.noise_pdf.rvs()
self.theta_noise = norm(scale=noise_std)
# bias #
self.bias_rate_nu = norm.rvs(loc=1.0, scale=bias_rate_stds[0])
self.bias_rate_omega = norm.rvs(loc=1.0, scale=bias_rate_stds[1])
# stuck #
self.stuck_pdf = expon(scale=expected_stuck_time)
self.escape_pdf = expon(scale=expected_escape_time)
self.time_until_stuck = self.stuck_pdf.rvs()
self.time_until_escape = self.escape_pdf.rvs()
self.is_stuck = False
# kidnap #
self.kidnap_pdf = expon(scale=expected_kidnap_time)
self.time_until_kidnap = self.kidnap_pdf.rvs()
rx, ry = kidnap_range_x, kidnap_range_y
self.kidnap_dist = uniform(loc=(rx[0], ry[0], 0.0), scale=(rx[1]-rx[0], ry[1]-ry[0], 2*math.pi))
def test_random_search_results():
# Make a dataset with a lot of noise to get various kind of prediction
# errors across CV folds and parameter settings
X, y = make_classification(n_samples=200, n_features=100, n_informative=3,
random_state=0)
# scipy.stats dists now supports `seed` but we still support scipy 0.12
# which doesn't support the seed. Hence the assertions in the test for
# random_search alone should not depend on randomization.
n_folds = 3
n_search_iter = 30
params = dict(C=expon(scale=10), gamma=expon(scale=0.1))
random_search = RandomizedSearchCV(SVC(), n_iter=n_search_iter, cv=n_folds,
iid=False, param_distributions=params)
random_search.fit(X, y)
random_search_iid = RandomizedSearchCV(SVC(), n_iter=n_search_iter,
cv=n_folds, iid=True,
param_distributions=params)
random_search_iid.fit(X, y)
param_keys = ('param_C', 'param_gamma')
score_keys = ('test_mean_score', 'test_rank_score',
'test_split0_score', 'test_split1_score',
'test_split2_score', 'test_std_score')
n_cand = n_search_iter
for search, iid in zip((random_search, random_search_iid), (False, True)):
assert_equal(iid, search.iid)
results = search.results_
# Check results structure
check_results_array_types(results, param_keys, score_keys)
check_results_keys(results, param_keys, score_keys, n_cand)
# For random_search, all the param array vals should be unmasked
assert_false(any(results['param_C'].mask) or
any(results['param_gamma'].mask))
check_results_grid_scores_consistency(search)
def tpdf(x, Ns, Nb, tlb, comps=["sig", "bkg"]):
sig = expon(trange[0], tlb)
sigN = np.diff(sig.cdf(trange))
bkg = uniform(trange[0], trange[1] - trange[0])
bkgN = np.diff(bkg.cdf(trange))
tot = 0
if "sig" in comps:
tot += Ns * sig.pdf(x) / sigN
if "bkg" in comps:
tot += Nb * bkg.pdf(x) / bkgN
return tot
def square_error_exp(_lambda):
from scipy.stats import expon
distribution = expon(scale=1 / _lambda)
square_errors = [
np.power(mean - distribution.mean(), 2.0) * mean_error_weight,
np.power(lejp - distribution.ppf(percentile_lower), 2.0) *
lejp_error_weight,
np.power(uejp - distribution.ppf(percentile_upper), 2.0) *
uejp_error_weight
]
return square_errors
def testExponentialSampleMultiDimensional(self):
with self.test_session():
batch_size = 2
lam_v = [3.0, 22.0]
lam = tf.constant([lam_v] * batch_size)
exponential = tf.contrib.distributions.Exponential(lam=lam)
n = 100000
samples = exponential.sample_n(n, seed=138)
self.assertEqual(samples.get_shape(), (n, batch_size, 2))
sample_values = samples.eval()
self.assertFalse(np.any(sample_values < 0.0))
for i in range(2):
self.assertLess(
stats.kstest(sample_values[:, 0, i],
stats.expon(scale=1.0 / lam_v[i]).cdf)[0],
0.01)
self.assertLess(
stats.kstest(sample_values[:, 1, i],
stats.expon(scale=1.0 / lam_v[i]).cdf)[0],
0.01)
def __init__(self,
t,
mu_s,
sigma_s,
mu,
stot,
btot,
bin_edges,
use_gaussian_appoximation=False):
# render all params positive
t, mu_s, sigma_s, mu, stot, btot = np.abs(
[t, mu_s, sigma_s, mu, stot, btot])
stot, btot = int(stot), int(btot)
self.t, self.mu_s, self.sigma_s, self.mu, self.stot, self.btot = t, mu_s, sigma_s, mu, stot, btot
self.bin_edges = bin_edges
self.n_bins = len(bin_edges) - 1
self.params = {
'stot': self.stot,
'btot': self.btot,
't': self.t,
'mu_s': self.mu_s,
'sigma_s': self.sigma_s,
"mu": self.mu
}
# base continuous distributions
self.b = stats.expon(scale=1 / t)
self.s = stats.norm(loc=mu_s, scale=sigma_s)
# compute distribution for each bin
self.si = self.stot * pd.Series(
[self.s.cdf(i) for i in self.bin_edges]).diff().dropna().values
self.bi = self.btot * pd.Series(
[self.b.cdf(i) for i in self.bin_edges]).diff().dropna().values
if len(self.si) != len(self.bi):
print("ERROR!!", self.t, self.mu_s, self.sigma_s, mu, btot, stot)
self.ni = self.mu * self.si + self.bi
if not use_gaussian_appoximation:
self.bins_distributions = [
stats.poisson(mu=self.ni[i]).pmf for i in range(len(self.ni))
]
else:
self.bins_distributions = [
stats.norm(loc=self.ni[i],
scale=np.sqrt(self.ni[i]) + 1e-50).pdf
for i in range(len(self.ni))
]