def __init__(self, concentration):
# Check input
ensure_tensor_like(concentration, "concentration")
# Store args
self.concentration = concentration
def __init__(self, distributions, logits=None, probs=None):
# Check input
if logits is None and probs is None:
raise ValueError("must pass either logits or probs")
if probs is not None:
ensure_tensor_like(probs, "probs")
if logits is not None:
ensure_tensor_like(logits, "logits")
# Distributions should be a pf, tf, or pt distribution
if not isinstance(distributions, BaseDistribution):
if get_backend() == "pytorch":
import torch.distributions as tod
if not isinstance(distributions, tod.Distribution):
raise TypeError(
"requires either a ProbFlow or PyTorch distribution")
else:
from tensorflow_probability import distributions as tfd
if not isinstance(distributions, tfd.Distribution):
raise TypeError(
"requires either a ProbFlow or TensorFlow distribution"
)
# Store args
self.distributions = distributions
self.logits = logits
self.probs = probs
def __init__(self, rate):
# Check input
ensure_tensor_like(rate, "rate")
# Store args
self.rate = rate
def __init__(self, concentration, scale):
# Check input
ensure_tensor_like(concentration, "concentration")
ensure_tensor_like(scale, "scale")
# Store args
self.concentration = concentration
self.scale = scale
def __init__(self, loc, cov):
# Check input
ensure_tensor_like(loc, "loc")
ensure_tensor_like(cov, "cov")
# Store args
self.loc = loc
self.cov = cov
def __init__(self, loc=0, scale=1):
# Check input
ensure_tensor_like(loc, "loc")
ensure_tensor_like(scale, "scale")
# Store args
self.loc = loc
self.scale = scale
def __init__(self, concentration, rate):
# Check input
ensure_tensor_like(concentration, "concentration")
ensure_tensor_like(rate, "rate")
# Store args
self.concentration = concentration
self.rate = rate
def __init__(self, logits=None, probs=None):
# Check input
if logits is None and probs is None:
raise TypeError("either logits or probs must be specified")
if logits is None:
ensure_tensor_like(probs, "probs")
if probs is None:
ensure_tensor_like(logits, "logits")
# Store args
self.logits = logits
self.probs = probs
def __init__(self, distributions, logits=None, probs=None):
# Check input
# TODO: distributions should be a pf, tf, or pt distribution
if logits is None and probs is None:
raise ValueError("must pass either logits or probs")
if probs is not None:
ensure_tensor_like(probs, "probs")
if logits is not None:
ensure_tensor_like(logits, "logits")
# Store args
self.distributions = distributions
self.logits = logits
self.probs = probs
def __init__(self, initial, transition, observation, steps):
# Check input
ensure_tensor_like(initial, "initial")
ensure_tensor_like(transition, "transition")
# observation should be a pf, tf, or pt distribution
if not isinstance(steps, int):
raise TypeError("steps must be an int")
if steps < 1:
raise ValueError("steps must be >0")
# Store observation distribution
if isinstance(observation, BaseDistribution):
self.observation = observation() # store backend distribution
else:
self.observation = observation
# Store other args
self.initial = initial
self.transition = transition
self.steps = steps