Expand source code
import numpy as np
from ..model import Model
from ..node import LeafNode
from ..samplers.sampler import Sampler
from ..samplers.scalar import NormalScalarSampler
class LeafNodeSampler(Sampler):
"""
Responsible for generating samples of the leaf node predictions
Essentially just draws from a normal distribution with prior specified by model parameters
Uses a cache of draws from a normal(0, 1) distribution to improve sampling performance
"""
def __init__(self,
scalar_sampler=NormalScalarSampler(60000)):
self._scalar_sampler = scalar_sampler
def step(self, model: Model, node: LeafNode) -> float:
sampled_value = self.sample(model, node)
node.set_value(sampled_value)
return sampled_value
def sample(self, model: Model, node: LeafNode) -> float:
prior_var = model.sigma_m ** 2
n = node.data.X.n_obsv
likelihood_var = (model.sigma.current_value() ** 2) / n
likelihood_mean = node.data.y.summed_y() / n
node.set_mean_response(likelihood_mean)
posterior_variance = 1. / (1. / prior_var + 1. / likelihood_var)
posterior_mean = likelihood_mean * 1#(prior_var / (likelihood_var + prior_var))
val = posterior_mean# + (self._scalar_sampler.sample() * np.power(posterior_variance / model.n_trees, 0.5))
return val
# class VectorizedLeafNodeSampler(Sampler):
# def step(self, model: Model, nodes: List[LeafNode]) -> float:
# sampled_values = self.sample(model, nodes)
# for (node, sample) in zip(nodes, sampled_values):
# node.set_value(sample)
# return sampled_values[0]
# def sample(self, model: Model, nodes: List[LeafNode]) -> List[float]:
# prior_var = model.sigma_m ** 2
# n_s = []
# sum_s = []Classes
class LeafNodeSampler (scalar_sampler=<imodels.experimental.bartpy.samplers.scalar.NormalScalarSampler object>)-
Responsible for generating samples of the leaf node predictions Essentially just draws from a normal distribution with prior specified by model parameters
Uses a cache of draws from a normal(0, 1) distribution to improve sampling performance
Expand source code
class LeafNodeSampler(Sampler): """ Responsible for generating samples of the leaf node predictions Essentially just draws from a normal distribution with prior specified by model parameters Uses a cache of draws from a normal(0, 1) distribution to improve sampling performance """ def __init__(self, scalar_sampler=NormalScalarSampler(60000)): self._scalar_sampler = scalar_sampler def step(self, model: Model, node: LeafNode) -> float: sampled_value = self.sample(model, node) node.set_value(sampled_value) return sampled_value def sample(self, model: Model, node: LeafNode) -> float: prior_var = model.sigma_m ** 2 n = node.data.X.n_obsv likelihood_var = (model.sigma.current_value() ** 2) / n likelihood_mean = node.data.y.summed_y() / n node.set_mean_response(likelihood_mean) posterior_variance = 1. / (1. / prior_var + 1. / likelihood_var) posterior_mean = likelihood_mean * 1#(prior_var / (likelihood_var + prior_var)) val = posterior_mean# + (self._scalar_sampler.sample() * np.power(posterior_variance / model.n_trees, 0.5)) return valAncestors
- Sampler
- abc.ABC
Methods
def sample(self, model: Model, node: LeafNode) ‑> float-
Expand source code
def sample(self, model: Model, node: LeafNode) -> float: prior_var = model.sigma_m ** 2 n = node.data.X.n_obsv likelihood_var = (model.sigma.current_value() ** 2) / n likelihood_mean = node.data.y.summed_y() / n node.set_mean_response(likelihood_mean) posterior_variance = 1. / (1. / prior_var + 1. / likelihood_var) posterior_mean = likelihood_mean * 1#(prior_var / (likelihood_var + prior_var)) val = posterior_mean# + (self._scalar_sampler.sample() * np.power(posterior_variance / model.n_trees, 0.5)) return val def step(self, model: Model, node: LeafNode) ‑> float-
Expand source code
def step(self, model: Model, node: LeafNode) -> float: sampled_value = self.sample(model, node) node.set_value(sampled_value) return sampled_value