def generate_doc(i, phi, theta, num_words_in_doc=WORDS_IN_DOC):
"i - порядковый номер генерируемого документа. Не должен превышать число столбцов theta
topicvec = multinomial.rvs(num_words_in_doc, theta[:,i], size = 1, random_state = 1)
words_in_doc = np.zeros(len(phi))
for j in range(len(topicvec[0])):
words_in_doc = words_in_doc + multinomial.rvs(topicvec[0][j], phi[:,j], size = 1, random_state = 1)
return words_in_doc[0]
def markov_sequence(p_init: np.array, p_transition: np.array, sequence_length: int) -> List[int]:
"""
Generate a Markov sequence based on p_init and p_transition.
"""
if p_init is None:
p_init = equilibrium_distribution(p_transition)
initial_state = list(multinomial.rvs(1, p_init)).index(1)
states = [initial_state]
for _ in range(sequence_length - 1):
p_tr = p_transition[states[-1]]
new_state = list(multinomial.rvs(1, p_tr)).index(1)
states.append(new_state)
return states
def multinomial(p, N, seed=None):
"""Multinomial distribution for a given probability p and total number of draws"""
if seed is not None:
np.random.seed(seed)
if np.sum(p) != 1:
p = np.array(p) / np.sum(p)
from scipy.stats import multinomial
if isinstance(N, (list, np.ndarray, pd.Series)):
return np.array([multinomial.rvs(n=int(n), p=p) for n in N])
elif is_numeric(N):
return multinomial.rvs(n=int(N), p=p)
else:
logging.error("Unknown N type", error_type=TypeError)
def _upsample_mortality(years=None, regions=None):
"""Returns deaggregated (per-case) mortality for density plots.
Args:
years (list, optional): List with years to contain. All by default.
regions (list, optional): List with years to contain. All by default.
Returns:
(pandas.DataFrame): Upsampled per-case mortality data.
"""
# get data
x = data()
# filter
if regions is not None:
x = x[x.region.isin(regions)]
if years is not None:
x = x[x.year.isin(years)]
# upsample
cases = {'sex': [], 'age': [], 'country': [], 'year': []}
for row in x.itertuples():
age_cat = row.age_end - row.age_start + 1
random_deaths = multinomial.rvs(int(row.deaths / 10), [1 / age_cat] *
age_cat) #, random_state = 12345)
ages = list(range(row.age_start, row.age_end + 1))
for age, deaths in zip(ages, random_deaths):
for _ in range(deaths):
cases['country'].append(row.region)
cases['year'].append(row.year)
cases['sex'].append(row.sex)
cases['age'].append(age)
cases = pd.DataFrame(cases)\
.sort_values(by = 'sex', ascending = False)
cases['date'] = None
# return
return cases
def generate(self, N=1):
"""Matches Language.generate: Generate 1 sample of N draws from theta.
Returned sample consists of an array size K with (integer) number of
draws of each category."""
norm_alpha = [a/sum(self.alpha) for a in self.alpha]
return multinomial.rvs(N, norm_alpha)
def compute_z_galaxies(q, N):
q[q <
0] = 0 # Quick fix: sometimes the number returned by the gaussian is negative and then the code crashes
p = q / q.sum()
draws = multinomial.rvs(n=1, p=p, size=N)
z = np.where(draws == 1)[1]
return z, np.stack(draws)
def rvs(self, size=None):
if size is None:
size = self.parameters.initial_infectious
assert (size >= self.parameters.initial_infectious)
# Loop until we get a satisfactory sample.
while True:
# Pick `size` random ages.
ages = self.age_structureRV.rvs(size=size)
# Determine the status for each age.
proportions = self._proportion(ages)
status = proportions.columns
status_ages = {k: [] for k in status}
# `scipy.stats.multinomial.rvs()` can't handle multiple `p`s,
# so we need to loop.
for (age, row) in proportions.iterrows():
# Randomly pick a status.
rv = multinomial.rvs(1, row)
# `rv` is an array with `1` in the position
# picked and `0`s in the remaining positions.
# Convert that to the name.
s = status[rv == 1][0]
# Add this `age` to the status list.
status_ages[s].append(age)
if (len(status_ages['susceptible']) <
self.parameters.initial_infectious):
# We don't have enough susceptibles. Loop again.
continue
else:
# Convert a few susceptibles to infectious.
for _ in range(self.parameters.initial_infectious):
age = status_ages['susceptible'].pop()
status_ages['infectious'].append(age)
# This is a satisfactory sample, so end loop.
break
return status_ages
def test_accumulator():
"""
Tests that the posterior probability computed sequentially
via accumulation is equal to the posterior probability
computed in a batch manner.
"""
theta = np.array([1 / 3, 1 / 3, 1 / 3])
dirichlet_probability = np.array([1, 3, 2])
dirichlet_concentration = 1
dirichlet_alpha = dirichlet_probability * dirichlet_concentration
sample_size = 40
observations = multinomial.rvs(1, theta, size=sample_size)
observations_sum = reduce(lambda x, y: x + y, observations)
final_posterior = sequential_posteriors(
observations,
theta,
dirichlet_probability=dirichlet_probability,
dirichlet_concentration=dirichlet_concentration,
)[-1]
final_bf = bayes_factor(final_posterior)
post_prob = posterior_probability(final_bf)
log_marginal_likelihood_M1 = log_posterior_predictive(
observations_sum, dirichlet_alpha)
log_marginal_likelihood_M0 = multinomial.logpmf(observations_sum,
observations_sum.sum(),
theta)
log_odds = log_marginal_likelihood_M1 - log_marginal_likelihood_M0
odds = np.exp(log_odds)
assert post_prob == approx(odds / (1 + odds))
def multinomial_sample(X, lam, rng=None):
"""
This draws multinomial samples from an urn using some poisson
process denoted by lam.
Parameters
----------
X: array_like
A matrix of counts where there are `n` rows and `m` columns
where `n` corresponds to the number of samples and `m`
corresponds to the number of species.
lam : float
Poisson parameter, which is also the mean and variance
of the Poisson.
rng: np.random.RandomState
Numpy random state number generator.
Returns
-------
np.array:
A matrix of counts where
there are `n` rows and `m` columns where `n` corresponds
to the number of samples and `m` corresponds to the number
of species.
"""
if rng is None:
rng = RandomState(0)
seq_depths = poisson.rvs(lam, size=X.shape[0], random_state=rng)
counts = [
multinomial.rvs(seq_depths[i], X[i, :], random_state=rng)
for i in range(len(seq_depths))
]
return np.vstack(counts)
def plot_violin(save = False, name = 'img/demographic/population.png'):
"""Constructs a violin plot of population (so called demographic curve).
Args:
save (bool, optional): Whether to cache or not.
name (str, optional): Path of caching.
"""
# fetch data
df = _populations_data()
# upsample
cases = {'sex': [], 'age': [], 'country': []}
for row in df.itertuples():
age_cat = row.age_end - row.age_start + 1
random_pops = multinomial.rvs(int(row.population / 100), [1/age_cat]*age_cat)
ages = list(range(row.age_start, row.age_end + 1))
for age,deaths in zip(ages, random_pops):
for _ in range(deaths):
cases['country'].append(row.region)
cases['sex'].append(row.sex)
cases['age'].append(age)
cases = pd.DataFrame(cases)\
.sort_values(by = 'sex', ascending = False)
cases['date'] = None
# plot
fig1, ax1 = plt.subplots()
sns.violinplot(x="country", y="age", hue="sex", data = cases, ax=ax1)
if save: fig1.savefig(name)
def sample(self, point, n_samples=1):
"""Sample from the categorical distribution.
Sample from the categorical distribution with parameters provided by
point. This gives samples in the simplex.
Parameters
----------
point : array-like, shape=[..., dim + 1]
Parameters of a categorical distribution, i.e. probabilities
associated to dim + 1 outcomes.
n_samples : int
Number of points to sample with each set of parameters in point.
Optional, default: 1.
Returns
-------
samples : array-like, shape=[..., n_samples]
Samples from categorical distributions.
"""
geomstats.errors.check_belongs(point, self)
point = gs.to_ndarray(point, to_ndim=2)
samples = []
for param in point:
counts = multinomial.rvs(1, param, size=n_samples)
samples.append(gs.argmax(counts, axis=-1))
return samples[0] if len(point) == 1 else gs.stack(samples)
def multinomial_sample(X, depths, rng=None):
"""
This draws multinomial samples from an urn using some poisson
process denoted by lam.
Parameters
----------
X: array_like
A matrix of counts where there are `n` rows and `m` columns
where `n` corresponds to the number of samples and `m`
corresponds to the number of species.
depths : np.array
Sampling depths for each of the multinomial samples.
rng: np.random.RandomState
Numpy random state number generator.
Returns
-------
np.array:
A matrix of counts where
there are `n` rows and `m` columns where `n` corresponds
to the number of samples and `m` corresponds to the number
of species.
"""
if rng is None:
rng = RandomState(0)
counts = [
multinomial.rvs(depths[i], X[i, :], random_state=rng)
for i in range(len(depths))
]
return np.vstack(counts)
def draw(self, K = 10, N = 1*10**5, m = 3, gaussian = False):
if self.seed is not None:
np.random.seed(self.seed)
alphas = gamma.rvs(5, size=m) # shape parameter
#print(sum(alphas)) # equivalent sample size
self.p = dirichlet.rvs(alpha = alphas, size = 1)[0]
self.phi_is = multinomial.rvs(1, self.p, size=N) # draw from categorical p.m.f
self.x_draws = np.zeros((N,K))
self.hyper_loc, self.hyper_scale, self.thetas, self.var, self.covs, self.rdraws = dict(), dict(), dict(), tuple(), tuple(), tuple()
for i in range(m):
self.hyper_loc["mean"+str(i+1)] = norm.rvs(size = 1, loc = 0, scale = 5)
self.hyper_scale["scale"+str(i+1)] = 1/gamma.rvs(5, size=1)
self.thetas["mean"+str(i+1)] = norm.rvs(size = K, loc = self.hyper_loc["mean"+str(i+1)],
scale = self.hyper_scale["scale"+str(i+1)])
self.thetas["Sigma"+str(i+1)] = np.eye(K)*(1/gamma.rvs(5, size=K))
self.thetas["nu"+str(i+1)] = randint.rvs(K+2, K+10, size=1)[0]
if gaussian:
self.covs += (self.thetas['Sigma'+str(i+1)], )
else:
self.covs += (wishart.rvs(df = self.thetas['nu'+str(i+1)], scale = self.thetas['Sigma'+str(i+1)], size=1),)
self.var += (self.thetas["nu"+str(i+1)]/(self.thetas["nu"+str(i+1)]-2)*self.covs[i],) # variance covariance matrix of first Student-t component
self.rdraws += (np.random.multivariate_normal(self.thetas["mean"+str(i+1)], self.covs[i], N),)
self.Phi = np.tile(self.phi_is[:,i], K).reshape(K,N).T # repeat phi vector to match with random matrix
self.x_draws += np.multiply(self.Phi, self.rdraws[i])
return self.x_draws
def gen_surrogate_data(n_point, p_cat, low, high, alpha, xmin, xmax, discrete,
random_state):
"""
Generate surrogate data points
:param n_point: total number of data points
:param p_cat: probability of `low`, `pareto` and `high` categories
:param low, high: data to be subsampled (with replacement) for categories `low` and `high`
:param alpha: exponent of the `pareto` regime
:param xmin, xmax: boundaries of the `pareto` regime, so that all(low<xmin) and all (xmax<=high)
:param discrete: use zipf distribution instead of pareto, bool
:param random_state:
:return: surrogate sample
"""
random_state = check_random_state(random_state)
s_low, s_mid, s_high = multinomial.rvs(n_point,
p_cat,
random_state=random_state)
sample = np.empty(n_point, dtype=float)
if s_low:
sample[0:s_low] = random_state.choice(low, s_low, replace=True)
if s_high:
sample[s_low + s_mid:n_point] = random_state.choice(high,
s_high,
replace=True)
sample[s_low:s_low + s_mid] = dispatch_rvs(alpha,
xmin,
xmax,
discrete,
size=s_mid,
random_state=random_state)
random_state.shuffle(sample)
return sample
def create_dataset(n_dim,
n_clust,
n_tasks,
n_entities,
seed=None,
pi_samp=None,
Si_samp=None,
mu_samp=None):
"""
Create the amortised clustering dataset
:param n_dim: number of dimensions
:param n_clust: pair (lo,hi) number of clusters uniformly in the range(lo,hi)
:param n_tasks: number of tasks
:param n_entities: pair (lo,hi) number of entities uniformly in the range(lo,hi)
:param seed: random seed
:return: data set
"""
if seed is not None:
np.random.seed(seed)
tasks = []
for i in range(n_tasks):
n_clust_ = np.random.randint(*n_clust)
Si = np.zeros((n_clust_, n_dim, n_dim))
mu = np.zeros((n_clust_, n_dim))
x = []
idx = []
n_ent = np.random.randint(*n_entities)
if pi_samp is not None:
pi = pi_samp(n_clust_)
else:
pi = np.ones(n_clust_) / n_clust_
for j, n in enumerate(*multinomial.rvs(n_ent, pi, 1)):
if Si_samp is not None:
Si[j] = Si_samp(n_dim)
else:
Si[j] = invwishart.rvs(4, 0.05 * np.eye(n_dim))
if mu_samp is not None:
mu[j] = mu_samp(n_dim)
else:
mu[j] = np.random.randn(n_dim)
if n > 0:
x.append(
multivariate_normal.rvs(mu[j], Si[j], size=[n]).astype(
np.float32).reshape(n, -1))
idx.append(j * np.ones(n, dtype=np.long))
j = np.random.permutation(n_ent)
x = np.concatenate(x, 0)[j]
idx = np.concatenate(idx, 0)[j]
tasks.append((x, idx, mu, Si))
return tasks
def draw(self, K=10, N=1 * 10**5, m=3, gaussian=False):
"""
Inputs:
-------
N: sample size
K: Dimension of Normal/Student distr.
m: number of mixture components
"""
np.random.seed(self.seed)
self.st0 = np.random.get_state() # get initial state of RNG
#np.random.set_state(self.st0)
print("Drawing from", m, "component mixture distribution.")
alphas = gamma.rvs(5, size=m) # shape parameter
#print(sum(alphas)) # equivalent sample size
self.p = dirichlet.rvs(alpha=alphas, size=1)[0]
self.phi_is = multinomial.rvs(1, self.p,
size=N) # draw from categorical p.m.f
self.x_draws = np.zeros((N, K))
self.hyper_loc, self.hyper_scale, self.thetas, self.var, self.covs, self.rdraws = dict(
), dict(), dict(), tuple(), tuple(), tuple()
for i in range(m):
self.hyper_loc["mean" + str(i + 1)] = norm.rvs(size=1,
loc=0,
scale=5)
self.hyper_scale["scale" + str(i + 1)] = 1 / gamma.rvs(5, size=1)
self.thetas["mean" + str(i + 1)] = norm.rvs(
size=K,
loc=self.hyper_loc["mean" + str(i + 1)],
scale=self.hyper_scale["scale" + str(i + 1)])
self.thetas["Sigma" +
str(i + 1)] = np.eye(K) * (1 / gamma.rvs(5, size=K))
self.thetas["nu" + str(i + 1)] = randint.rvs(K + 2, K + 10,
size=1)[0]
if gaussian:
self.covs += (self.thetas['Sigma' + str(i + 1)], )
else:
self.covs += (wishart.rvs(df=self.thetas['nu' + str(i + 1)],
scale=self.thetas['Sigma' +
str(i + 1)],
size=1), )
self.var += (
self.thetas["nu" + str(i + 1)] /
(self.thetas["nu" + str(i + 1)] - 2) * self.covs[i],
) # variance covariance matrix of first Student-t component
self.rdraws += (np.random.multivariate_normal(
self.thetas["mean" + str(i + 1)], self.covs[i], N), )
self.Phi = np.tile(self.phi_is[:, i], K).reshape(
K, N).T # repeat phi vector to match with random matrix
self.x_draws += np.multiply(self.Phi, self.rdraws[i])
return self.x_draws, np.argmax(self.phi_is, 1) # X, latent
def multinomial_robust(NN, p, size=None):
if NN < 1000:
return multinomial.rvs(NN, p)
else:
results = np.array([binomial_robust(NN, pi, size) for pi in p])
last_entry = int(NN) - results[:-1].sum(0)
while last_entry < 0:
results = np.array([binomial_robust(NN, pi, size) for pi in p])
last_entry = int(NN) - results[:-1].sum(0)
return np.rollaxis(results, 0, results.ndim)
def test_p_values_decreasing_and_in_range():
p_0 = np.array([1 / 3, 1 / 3, 1 / 3])
p_1 = np.array([2 / 9, 4 / 9, 3 / 9])
sample_size = 40
data = multinomial.rvs(1, p_1, size=sample_size)
pvals = sequential_p_values(data, p_0)
for ix in range(1, sample_size):
assert pvals[ix] <= pvals[ix - 1] # pvals should be non increasing
for pval in pvals:
assert 0.0 <= pval and pval <= 1.0
def samples(self, F, num_samples,Y_metadata=None):
eF = safe_exp(F)
den = 1 + eF.sum(1)[:, None]
p = eF / np.tile(den, eF.shape[1])
p = np.hstack((p, 1 / den))
p = np.clip(p, 1e-9, 1 - 1e-9)
p = p / np.tile(p.sum(1)[:,None], (1, p.shape[1]))
samples = np.empty((F.shape[0], self.K))
for i in range(F.shape[0]):
samples[i,:] = multinomial.rvs(n=1, p=p[i,:], size=1)
return self.invonehot(Y=samples)
def create_multinomial_doublet(X: np.ndarray, i: int, j: int, **kwargs):
'''make a multinomial combination of 2 cells
Parameters
----------
X : np.array
cell by genes matrix
i : int,
randomly chosen ith cell
j : int,
randomly chosen jth cell
kwargs : dict,
dict with doublet_depth, cell_depths and cells_ids as keys
doublet_depth is an int
cell_depths is an list of all cells total UMI counts as ints
cell_ids list of lists with genes with counts for each cell
Returns
-------
float
multinomial expression vector of two cells
'''
doublet_depth = kwargs["doublet_depth"]
cell_depths = kwargs["cell_depths"]
cells_ids = kwargs["cells_ids"]
randomize_doublet_size = kwargs["randomize_doublet_size"]
# add their counts
dp = X[i] + X[j]
non_zero_indexes = np.unique(cells_ids[i] + cells_ids[j])
if issparse(X):
dp = dp.data
else:
dp = np.ravel(dp)
dp = dp[non_zero_indexes]
# a huge hack caused by
# https://github.com/numpy/numpy/issues/8317
# fun fun fun https://stackoverflow.com/questions/23257587/how-can-i-avoid-value-errors-when-using-numpy-random-multinomial
# okay with this hack because affects pro
# normalize
dp /= dp.sum()
if randomize_doublet_size:
scale_factor = np.random.uniform(1., doublet_depth)
else:
scale_factor = doublet_depth
# choose depth
dd = int(scale_factor * (cell_depths[i] + cell_depths[j]) / 2)
# sample counts from multinomial
non_zero_probs = multinomial.rvs(n=dd, p=dp)
probs = np.zeros(X.shape[1])
probs[non_zero_indexes] = non_zero_probs
return csr_matrix(probs) if issparse(X) else probs
def transform_single(index):
column = X[:, index].copy()
mask = pd.isnull(column)
values, probabilities = self.statistics_[index]
sample = np.argmax(multinomial.rvs(p=probabilities,
n=1,
size=mask.sum(),
random_state=self.random_state),
axis=1)
column[mask] = np.vectorize(lambda pick: values[pick])(sample)
return column
def sample_from_multinomial(self, sampletimes=1):
if self.valuenumber is None:
self.set_random_numbers()
self.set_random_probabilities()
sample = np.zeros(sampletimes)
for x in xrange(0, sampletimes):
One_sample = multinomial.rvs(1, self.probabilities)
sample[x] = np.where(One_sample == 1)[0] + 1
self.temporary_sample = sample
def sample(self, size=1):
import numpy as np
if isinstance(size, int):
if self.input is None: size = [0]*size
else: size = self.input.sample(size)
elif self.input is None: raise ValueError('no input model provided to index into')
params = [self.param(self.get_beta(idx)) for idx in size]
if self.kind=='con':
from scipy.stats import norm
return np.array([norm.rvs(loc=p, scale=self.sigma, size=1) for p in params])
else:
from scipy.stats import multinomial
return np.array([multinomial.rvs(n=1, p=p, size=1).argmax() for p in params])
def __getitem__(self, index):
indexes = self.indexes[index * self.batch_size:(index + 1) *
self.batch_size]
X = self.X_train[indexes, :, :, :]
if self.is_data_augment and not self.is_validate:
if self.data_augment_noise_type == 'normal':
noises = np.random.normal(loc=self.normal_loc,
scale=1.,
size=X.shape)
elif self.data_augment_noise_type == 'uniform':
noises = np.random.uniform(low=-1., high=1., size=X.shape)
X += 0.001 * noises
if self.is_validate:
y = self.y_train[indexes, :]
else:
num_label_case = np.sum([
self.is_soft_label, self.is_sample_label_dist,
self.is_mix_label_original
])
assert num_label_case == len(self.list_label_case)
idx_label_case = np.random.randint(num_label_case,
size=len(indexes))
y = np.zeros((len(indexes), self.y_train.shape[1]))
for i, ind in enumerate(indexes):
name_label_case = self.list_label_case[idx_label_case[i]]
if name_label_case == 'soft_label':
y[i] = self.y_train[ind]
elif name_label_case == 'sample_label_dist':
y_prob = self.y_train[ind].astype(np.float64)
y_prob /= np.sum(y_prob)
if np.sum(y_prob) > 1: # due to numerical precision
y_prob += np.finfo(float).eps
y_prob /= np.sum(y_prob)
y[i] = multinomial.rvs(1, y_prob, 1)
elif name_label_case == 'mix_label_original':
y[i] = self.y_original[ind]
else:
raise ValueError(
'not existing label case: {}'.format(name_label_case))
#print (i, ind, name_label_case,y[i])
assert np.sum(y) == len(indexes)
assert np.all(
np.abs(np.sum(y, axis=1) - 1.) < 1e-5
), 'label probability does not sum up to 1\n{}\n{}'.format(
np.sum(y, axis=1), y)
return X, y
def __call__(self, NUM=None, P=None, sampletimes=1):
self.NUM = NUM
if P is None:
if NUM is None:
self.gen_NUM()
self.gen_P()
else:
self.P = P
# Return a Sample
sample = np.zeros(sampletimes)
for x in xrange(0, sampletimes):
One_sample = multinomial.rvs(1, self.P)
sample[x] = np.where(One_sample == 1)[0]
return sample
def distribute_doses(self, model: SIR) -> Tuple[np.array]:
if self.exhausted(model):
return (np.zeros(self.age_ratios.shape),
np.zeros(self.age_ratios.shape),
np.zeros(self.age_ratios.shape))
dV = (model.S[-1] /
model.N[-1]) * self.daily_doses * self.effectiveness
model.S[-1] -= dV
model.parallel_forward_epi_step()
distributed_doses = Multinomial.rvs(self.daily_doses, self.age_ratios)
effective_doses = self.effectiveness * distributed_doses
immunizing_doses = (model.S[-1].mean() /
model.N[-1].mean()) * effective_doses
self.bin_populations -= immunizing_doses.astype(int)
return (distributed_doses, effective_doses, immunizing_doses)
def gen_surrogate_counts(n_point, p_cat, p_low, p_high, alpha, xmin, xmax,
bins, discrete, random_state):
"""
Generate surrogate hit counts
:param n_point: total number of data points
:param p_cat: probability of `low`, `pareto` and `high` categories
:param p_low, p_high: hit probabilities within categories `low` and `high`
:param alpha: exponent of the `pareto` regime
:param xmin, xmax: boundaries of the `pareto` regime, so that all(low<xmin) and all(xmax<=high)
:param bins: bin boundaries (used for calculating cdf and or binning samples)
:param discrete: use zipf distribution instead of pareto, bool
:param random_state:
:return: surrogate hit counts
"""
random_state = check_random_state(random_state)
s_low, s_mid, s_high = multinomial.rvs(n_point,
p_cat,
random_state=random_state)
# TODO: the same can be achieved by using the cdf and multinomial sampling, see whether it is stable enough.
sample = dispatch_rvs(alpha,
xmin,
xmax,
discrete,
size=s_mid,
random_state=random_state)
counts, _ = np.histogram(sample, bins)
if s_low:
counts[0:len(p_low)] = multinomial.rvs(s_low,
p_low,
random_state=random_state)
if s_high:
counts[len(counts) - len(p_high):len(counts)] = multinomial.rvs(
s_low, p_high, random_state=random_state)
return counts
def _resample(self, n, prob, classes, grouped_data):
samples_no = multinomial.rvs(n=n,
p=prob,
random_state=self.random_state)
subset_x, subset_y = [], []
for no, j in enumerate(classes):
data = grouped_data[j]
resample_class = resample(data,
replace=True,
n_samples=samples_no[no],
random_state=self.random_state)
for sample in resample_class:
subset_x.append(sample[0])
subset_y.append(sample[1])
return np.array(subset_x), np.array(subset_y)
def compute_conditional_z(q, y, mu,
sigma_square): #Could be optimised in future work
n = np.shape(y)[0]
d = np.shape(mu)[0]
z = np.empty(shape=(n))
i = np.empty(shape=(n, d))
for l in range(n):
temp = np.empty(shape=(d))
for j in range(d):
temp[j] = q[j] * multivariate_normal.pdf(
y[l, 0], mean=mu[j], cov=sigma_square[j])
temp[temp < 0] = 0
temp = temp / np.sum(temp)
i[l, :] = multinomial.rvs(n=1, p=temp, size=1)[0]
z = np.where(i == 1)[1]
return z, i
def generate_data(n, seed=None, x=None):
if seed is not None:
np.random.seed(seed)
if x is None:
x = np.random.uniform(size=(n, 1))
eta_1, eta_2, eta_3 = etas(x)
class_probs = np.hstack((eta_1, eta_2, eta_3))
y_cats = np.array([
multinomial.rvs(1,
class_probs[i],
random_state=(seed if i == 0 else None))
for i in range(x.shape[0])
])
y = np.argmax(y_cats, axis=1)
return x, y