from convoys import autograd_scipy_monkeypatch # NOQA
import autograd
from autograd_gamma import gammainc
from deprecated.sphinx import deprecated
import emcee
import numpy
from scipy.special import gammaincinv
from autograd.scipy.special import expit, gammaln
from autograd.numpy import isnan, exp, dot, log, sum
import progressbar
import scipy.optimize
import warnings
__all__ = ['Exponential',
'Weibull',
'Gamma',
'GeneralizedGamma']
def generalized_gamma_loss(x, X, B, T, W, fix_k, fix_p,
hierarchical, flavor, callback=None):
k = exp(x[0]) if fix_k is None else fix_k
p = exp(x[1]) if fix_p is None else fix_p
log_sigma_alpha = x[2]
log_sigma_beta = x[3]
a = x[4]
b = x[5]
n_features = int((len(x)-6)/2)
alpha = x[6:6+n_features]
beta = x[6+n_features:6+2*n_features]
lambd = exp(dot(X, alpha)+a)
# PDF: p*lambda^(k*p) / gamma(k) * t^(k*p-1) * exp(-(x*lambda)^p)
log_pdf = log(p) + (k*p) * log(lambd) - gammaln(k) \
+ (k*p-1) * log(T) - (T*lambd)**p
cdf = gammainc(k, (T*lambd)**p)
if flavor == 'logistic': # Log-likelihood with sigmoid
c = expit(dot(X, beta)+b)
LL_observed = log(c) + log_pdf
LL_censored = log((1 - c) + c * (1 - cdf))
elif flavor == 'linear': # L2 loss, linear
c = dot(X, beta)+b
LL_observed = -(1 - c)**2 + log_pdf
LL_censored = -(c*cdf)**2
LL_data = sum(
W * B * LL_observed +
W * (1 - B) * LL_censored, 0)
if hierarchical:
# Hierarchical model with sigmas ~ invgamma(1, 1)
LL_prior_a = -4*log_sigma_alpha - 1/exp(log_sigma_alpha)**2 \
- dot(alpha, alpha) / (2*exp(log_sigma_alpha)**2) \
- n_features*log_sigma_alpha
LL_prior_b = -4*log_sigma_beta - 1/exp(log_sigma_beta)**2 \
- dot(beta, beta) / (2*exp(log_sigma_beta)**2) \
- n_features*log_sigma_beta
LL = LL_prior_a + LL_prior_b + LL_data
else:
LL = LL_data
if isnan(LL):
return -numpy.inf
if callback is not None:
callback(LL)
return LL
class RegressionModel(object):
pass
class GeneralizedGamma(RegressionModel):
''' Generalization of Gamma, Weibull, and Exponential
:param mcmc: boolean, defaults to False. Whether to use MCMC to
sample from the posterior so that a confidence interval can be
estimated later (see :meth:`predict`).
:param hierarchical: boolean denoting whether we have a (Normal) prior
on the alpha and beta parameters to regularize. The variance of
the normal distribution is in itself assumed to be an inverse
gamma distribution (1, 1).
:param flavor: defaults to logistic. If set to 'linear', then an
linear model is fit, where the beta params will be completely
additive. This creates a much more interpretable model, with some
minor loss of accuracy.
:param ci: boolean, deprecated alias for `mcmc`.
This mostly follows the `Wikipedia article
<https://en.wikipedia.org/wiki/Generalized_gamma_distribution>`_, although
our notation is slightly different. Also see `this paper
<http://data.princeton.edu/pop509/ParametricSurvival.pdf>`_ for an overview.
**Shape of the probability function**
The cumulative density function is:
:math:`F(t) = P(k, (t\\lambda)^p)`
where :math:`P(a, x) = \\gamma(a, x) / \\Gamma(a)` is the lower regularized
incomplete gamma function.
:math:`\\gamma(a, x)` is the incomplete gamma function and :math:`\\Gamma(a)`
is the standard gamma function.
The probability density function is:
:math:`f(t) = p\\lambda^{kp} t^{kp-1} \\exp(-(t\\lambda)^p) / \\Gamma(k)`
**Modeling conversion rate**
Since our goal is to model the conversion rate, we assume the conversion
rate converges to a final value
:math:`c = \\sigma(\\mathbf{\\beta^Tx} + b)`
where :math:`\\sigma(z) = 1/(1+e^{-z})` is the sigmoid function,
:math:`\\mathbf{\\beta}` is an unknown vector we are solving for (with
corresponding intercept :math:`b`), and :math:`\\mathbf{x}` are the
feature vector (inputs).
We also assume that the rate parameter :math:`\\lambda` is determined by
:math:`\\lambda = exp(\\mathbf{\\alpha^Tx} + a)`
where :math:`\\mathrm{\\alpha}` is another unknown vector we are
trying to solve for (with corresponding intercept :math:`a`).
We also assume that the :math:`\\mathbf{\\alpha}, \\mathbf{\\beta}`
vectors have a normal distribution
:math:`\\alpha_i \\sim \\mathcal{N}(0, \\sigma_{\\alpha})`,
:math:`\\beta_i \\sim \\mathcal{N}(0, \\sigma_{\\beta})`
where hyperparameters :math:`\\sigma_{\\alpha}^2, \\sigma_{\\beta}^2`
are drawn from an inverse gamma distribution
:math:`\\sigma_{\\alpha}^2 \\sim \\text{inv-gamma}(1, 1)`,
:math:`\\sigma_{\\beta}^2 \\sim \\text{inv-gamma}(1, 1)`
**List of parameters**
The full model fits vectors :math:`\\mathbf{\\alpha, \\beta}` and scalars
:math:`a, b, k, p, \\sigma_{\\alpha}, \\sigma_{\\beta}`.
**Likelihood and censorship**
For entries that convert, the contribution to the likelihood is simply
the probability density given by the probability distribution function
:math:`f(t)` times the final conversion rate :math:`c`.
For entries that *did not* convert, there is two options. Either the
entry will never convert, which has probability :math:`1-c`. Or,
it will convert at some later point that we have not observed yet,
with probability given by the cumulative density function
:math:`F(t)`.
**Solving the optimization problem**
To find the MAP (max a posteriori), `scipy.optimize.minimize
<https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.minimize.html#scipy.optimize.minimize>`_
with the SLSQP method.
If `mcmc == True`, then `emcee <http://dfm.io/emcee/current/>`_ is used
to sample from the full posterior in order to generate uncertainty
estimates for all parameters.
'''
def __init__(self, mcmc=False, fix_k=None, fix_p=None, hierarchical=True,
flavor='logistic', ci=None):
self._mcmc = mcmc
self._fix_k = fix_k
self._fix_p = fix_p
self._hierarchical = hierarchical
self._flavor = flavor
if ci is not None:
warnings.warn('The `ci` argument is deprecated in 0.2.1 in favor '
' of `mcmc`.', DeprecationWarning)
self._mcmc = ci
def fit(self, X, B, T, W=None):
'''Fits the model.
:param X: numpy matrix of shape :math:`k \\cdot n`
:param B: numpy vector of shape :math:`n`
:param T: numpy vector of shape :math:`n`
:param W: (optional) numpy vector of shape :math:`n`
'''
if W is None:
W = numpy.ones(len(X))
X, B, T, W = (Z if type(Z) == numpy.ndarray else numpy.array(Z)
for Z in (X, B, T, W))
keep_indexes = (T > 0) & (B >= 0) & (B <= 1) & (W >= 0)
if sum(keep_indexes) < X.shape[0]:
n_removed = X.shape[0] - sum(keep_indexes)
warnings.warn('Warning! Removed %d/%d entries from inputs where '
'T <= 0 or B not 0/1 or W < 0' % (n_removed, len(X)))
X, B, T, W = (Z[keep_indexes] for Z in (X, B, T, W))
n_features = X.shape[1]
# scipy.optimize and emcee forces the the parameters to be a vector:
# (log k, log p, log sigma_alpha, log sigma_beta,
# a, b, alpha_1...alpha_k, beta_1...beta_k)
# Generalized Gamma is a bit sensitive to the starting point!
x0 = numpy.zeros(6+2*n_features)
x0[0] = +1 if self._fix_k is None else log(self._fix_k)
x0[1] = -1 if self._fix_p is None else log(self._fix_p)
args = (X, B, T, W, self._fix_k, self._fix_p,
self._hierarchical, self._flavor)
# Set up progressbar and callback
bar = progressbar.ProgressBar(widgets=[
progressbar.Variable('loss', width=15, precision=9), ' ',
progressbar.BouncingBar(), ' ',
progressbar.Counter(width=6),
' [', progressbar.Timer(), ']'])
def callback(LL, value_history=[]):
value_history.append(LL)
bar.update(len(value_history), loss=LL)
# Define objective and use automatic differentiation
f = lambda x: -generalized_gamma_loss(x, *args, callback=callback)
jac = autograd.grad(lambda x: -generalized_gamma_loss(x, *args))
# Find the maximum a posteriori of the distribution
res = scipy.optimize.minimize(f, x0, jac=jac, method='SLSQP',
options={'maxiter': 9999})
if not res.success:
raise Exception('Optimization failed with message: %s' %
res.message)
result = {'map': res.x}
# TODO: should not use fixed k/p as search parameters
if self._fix_k:
result['map'][0] = log(self._fix_k)
if self._fix_p:
result['map'][1] = log(self._fix_p)
# Make sure we're in a local minimum
gradient = jac(result['map'])
gradient_norm = numpy.dot(gradient, gradient)
if gradient_norm >= 1e-2 * len(X):
warnings.warn('Might not have found a local minimum! '
'Norm of gradient is %f' % gradient_norm)
# Let's sample from the posterior to compute uncertainties
if self._mcmc:
dim, = res.x.shape
n_walkers = 5*dim
sampler = emcee.EnsembleSampler(
nwalkers=n_walkers,
ndim=dim,
log_prob_fn=generalized_gamma_loss,
args=args,
)
mcmc_initial_noise = 1e-3
p0 = [result['map'] + mcmc_initial_noise * numpy.random.randn(dim)
for i in range(n_walkers)]
n_burnin = 100
n_steps = numpy.ceil(2000. / n_walkers)
n_iterations = n_burnin + n_steps
bar = progressbar.ProgressBar(max_value=n_iterations, widgets=[
progressbar.Percentage(), ' ', progressbar.Bar(),
' %d walkers [' % n_walkers,
progressbar.AdaptiveETA(), ']'])
for i, _ in enumerate(sampler.sample(p0, iterations=n_iterations)):
bar.update(i+1)
result['samples'] = sampler.chain[:, n_burnin:, :] \
.reshape((-1, dim)).T
if self._fix_k:
result['samples'][0, :] = log(self._fix_k)
if self._fix_p:
result['samples'][1, :] = log(self._fix_p)
self.params = {k: {
'k': exp(data[0]),
'p': exp(data[1]),
'a': data[4],
'b': data[5],
'alpha': data[6:6+n_features].T,
'beta': data[6+n_features:6+2*n_features].T,
} for k, data in result.items()}
def _predict(self, params, x, t):
lambd = exp(dot(x, params['alpha'].T) + params['a'])
if self._flavor == 'logistic':
c = expit(dot(x, params['beta'].T) + params['b'])
elif self._flavor == 'linear':
c = dot(x, params['beta'].T) + params['b']
M = c * gammainc(
params['k'],
(t*lambd)**params['p'])
return M
def predict_posteriori(self, x, t):
''' Returns the trace samples generated via the MCMC steps.
Requires the model to be fit with `mcmc == True`.'''
x = numpy.array(x)
t = numpy.array(t)
assert self._mcmc
params = self.params['samples']
t = numpy.expand_dims(t, -1)
return self._predict(params, x, t)
def predict_ci(self, x, t, ci=0.8):
'''Works like :meth:`predict` but produces a confidence interval.
Requires the model to be fit with `ci = True`. The return value
will contain one more dimension than for :meth:`predict`, and
the last dimension will have size 3, containing the mean, the
lower bound of the confidence interval, and the upper bound of
the confidence interval.
'''
M = self.predict_posteriori(x, t)
y = numpy.mean(M, axis=-1)
y_lo = numpy.percentile(M, (1-ci)*50, axis=-1)
y_hi = numpy.percentile(M, (1+ci)*50, axis=-1)
return numpy.stack((y, y_lo, y_hi), axis=-1)
def predict(self, x, t):
'''Returns the value of the cumulative distribution function
for a fitted model (using the maximum a posteriori estimate).
:param x: feature vector (or matrix)
:param t: time
'''
params = self.params['map']
x = numpy.array(x)
t = numpy.array(t)
return self._predict(params, x, t)
def rvs(self, x, n_curves=1, n_samples=1, T=None):
''' Samples values from this distribution
T is optional and means we already observed non-conversion until T
'''
assert self._mcmc # Need to be fit with MCMC
if T is None:
T = numpy.zeros((n_curves, n_samples))
else:
assert T.shape == (n_curves, n_samples)
B = numpy.zeros((n_curves, n_samples), dtype=numpy.bool)
C = numpy.zeros((n_curves, n_samples))
params = self.params['samples']
for i, j in enumerate(numpy.random.randint(len(params['k']),
size=n_curves)):
k = params['k'][j]
p = params['p'][j]
lambd = exp(dot(x, params['alpha'][j]) + params['a'][j])
c = expit(dot(x, params['beta'][j]) + params['b'][j])
z = numpy.random.uniform(size=(n_samples,))
cdf_now = c * gammainc(
k,
numpy.multiply.outer(T[i], lambd)**p) # why is this outer?
adjusted_z = cdf_now + (1 - cdf_now) * z
B[i] = (adjusted_z < c)
y = adjusted_z / c
w = gammaincinv(k, y)
# x = (t * lambd)**p
C[i] = w**(1./p) / lambd
C[i][~B[i]] = 0
return B, C
@deprecated(version='0.2.0',
reason='Use :meth:`predict` or :meth:`predict_ci` instead.')
def cdf(self, x, t, ci=False):
'''Returns the predicted values.'''
if ci:
return self.predict_ci(x, t)
else:
return self.predict(x, t)
@deprecated(version='0.2.0',
reason='Use :meth:`predict_posteriori` instead.')
def cdf_posteriori(self, x, t):
'''Returns the a posterior distribution of the predicted values.'''
return self.predict_posteriori(x, t)
class Exponential(GeneralizedGamma):
''' Specialization of :class:`.GeneralizedGamma` where :math:`k=1, p=1`.
The cumulative density function is:
:math:`F(t) = 1 - \\exp(-t\\lambda)`
The probability density function is:
:math:`f(t) = \\lambda\\exp(-t\\lambda)`
The exponential distribution is the most simple distribution.
From a conversion perspective, you can interpret it as having
two competing final states where the probability of transitioning
from the initial state to converted or dead is constant.
See documentation for :class:`GeneralizedGamma`.'''
def __init__(self, *args, **kwargs):
kwargs.update(dict(fix_k=1, fix_p=1))
super(Exponential, self).__init__(*args, **kwargs)
class Weibull(GeneralizedGamma):
''' Specialization of :class:`.GeneralizedGamma` where :math:`k=1`.
The cumulative density function is:
:math:`F(t) = 1 - \\exp(-(t\\lambda)^p)`
The probability density function is:
:math:`f(t) = p\\lambda(t\\lambda)^{p-1}\\exp(-(t\\lambda)^p)`
See documentation for :class:`GeneralizedGamma`.'''
def __init__(self, *args, **kwargs):
kwargs.update(dict(fix_k=1))
super(Weibull, self).__init__(*args, **kwargs)
class Gamma(GeneralizedGamma):
''' Specialization of :class:`.GeneralizedGamma` where :math:`p=1`.
The cumulative density function is:
:math:`F(t) = P(k, t\\lambda)`
where :math:`P(a, x) = \\gamma(a, x) / \\Gamma(a)` is the lower regularized
incomplete gamma function.
The probability density function is:
:math:`f(t) = \\lambda^k t^{k-1} \\exp(-x\\lambda) / \\Gamma(k)`
See documentation for :class:`GeneralizedGamma`.'''
def __init__(self, *args, **kwargs):
kwargs.update(dict(fix_p=1))
super(Gamma, self).__init__(*args, **kwargs)
