def ap_loguniform_sampler(obs, prior_weight, low, high,
size=(), rng=None):
prior_mu = 0.5 * (high + low)
prior_sigma = 1.0 * (high - low)
weights, mus, sigmas = scope.adaptive_parzen_normal(
scope.log(obs), prior_weight, prior_mu, prior_sigma)
rval = scope.LGMM1(weights, mus, sigmas, low=low, high=high,
size=size, rng=rng)
return rval
def many_dists():
a = hp_choice('a', [0, 1, 2])
b = hp_randint('b', 10)
c = hp_uniform('c', 4, 7)
d = hp_loguniform('d', -2, 0)
e = hp_quniform('e', 0, 10, 3)
f = hp_qloguniform('f', 0, 3, 2)
g = hp_normal('g', 4, 7)
h = hp_lognormal('h', -2, 2)
i = hp_qnormal('i', 0, 10, 2)
j = hp_qlognormal('j', 0, 2, 1)
z = a + b + c + d + e + f + g + h + i + j
return {'loss': scope.float(scope.log(1e-12 + z**2))}
def many_dists():
a=hp_choice('a', [0, 1, 2])
b=hp_randint('b', 10)
c=hp_uniform('c', 4, 7)
d=hp_loguniform('d', -2, 0)
e=hp_quniform('e', 0, 10, 3)
f=hp_qloguniform('f', 0, 3, 2)
g=hp_normal('g', 4, 7)
h=hp_lognormal('h', -2, 2)
i=hp_qnormal('i', 0, 10, 2)
j=hp_qlognormal('j', 0, 2, 1)
z = a + b + c + d + e + f + g + h + i + j
return {'loss': scope.float(scope.log(1e-12 + z ** 2))}
def ap_qloguniform_sampler(obs, prior_weight, low, high, q, size=(), rng=None):
prior_mu = 0.5 * (high + low)
prior_sigma = 1.0 * (high - low)
weights, mus, sigmas = scope.adaptive_parzen_normal(
scope.log(
# -- map observations that were quantized to be below exp(low)
# (particularly 0) back up to exp(low) where they will
# interact in a reasonable way with the AdaptiveParzen
# thing.
scope.maximum(obs, scope.maximum(EPS, scope.exp(low))) # -- protect against exp(low) underflow
),
prior_weight,
prior_mu,
prior_sigma,
)
return scope.LGMM1(weights, mus, sigmas, low, high, q=q, size=size, rng=rng)
def many_dists():
a = hp_choice('a', [0, 1, 2])
b = hp_randint('b', 10)
c = hp_uniform('c', 4, 7)
d = hp_loguniform('d', -2, 0)
e = hp_quniform('e', 0, 10, 3)
f = hp_qloguniform('f', 0, 3, 2)
g = hp_normal('g', 4, 7)
h = hp_lognormal('h', -2, 2)
i = hp_qnormal('i', 0, 10, 2)
j = hp_qlognormal('j', 0, 2, 1)
k = hp_pchoice('k', [(.1, 0), (.9, 1)])
z = a + b + c + d + e + f + g + h + i + j + k
return {
'loss': scope.float(scope.log(1e-12 + z**2)),
'status': base.STATUS_OK
}
def ap_qloguniform_sampler(obs, prior_weight, low, high, q,
size=(), rng=None):
prior_mu = 0.5 * (high + low)
prior_sigma = 1.0 * (high - low)
weights, mus, sigmas = scope.adaptive_parzen_normal(
scope.log(
# -- map observations that were quantized to be below exp(low)
# (particularly 0) back up to exp(low) where they will
# interact in a reasonable way with the AdaptiveParzen
# thing.
scope.maximum(
obs,
scope.maximum( # -- protect against exp(low) underflow
EPS,
scope.exp(low)))),
prior_weight, prior_mu, prior_sigma)
return scope.LGMM1(weights, mus, sigmas, low, high, q=q,
size=size, rng=rng)
def ap_qlognormal_sampler(obs, prior_weight, mu, sigma, q, size=(), rng=None):
log_obs = scope.log(scope.maximum(obs, EPS))
weights, mus, sigmas = scope.adaptive_parzen_normal(
log_obs, prior_weight, mu, sigma)
rval = scope.LGMM1(weights, mus, sigmas, q=q, size=size, rng=rng)
return rval
def ap_loglognormal_sampler(obs, prior_weight, mu, sigma, size=(), rng=None):
weights, mus, sigmas = scope.adaptive_parzen_normal(
scope.log(obs), prior_weight, mu, sigma)
rval = scope.LGMM1(weights, mus, sigmas, size=size, rng=rng)
return rval
