Skip to content

Fix loglik and EDF for zero-inflated data #22

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
tnagler merged 5 commits into from
Aug 18, 2025
Merged

Fix loglik and EDF for zero-inflated data #22

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
ef80ad2
prevent /0 nans in edf computation
Jul 11, 2025
a8d0c9e
fix loglik for zero-inflated data (obs=0 was omitted)
Jul 11, 2025
0eda0f0
fix implicit cast
Jul 11, 2025
81860f2
safeguard weighted loglik computation
Jul 11, 2025
2dc6f31
add comment for loglik
Jul 21, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 21 additions & 6 deletions include/kde1d/kde1d.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -270,8 +270,9 @@ Kde1d::fit(const Eigen::VectorXd& x, const Eigen::VectorXd& weights)
Eigen::VectorXd w = weights;
tools::remove_nans(xx, w);

if (w.size() > 0)
if (w.size() > 0) {
w /= w.mean();
}

if (type_ == VarType::zero_inflated) {
if (w.size() == 0)
Expand Down Expand Up @@ -319,13 +320,24 @@ Kde1d::fit(const Eigen::VectorXd& x, const Eigen::VectorXd& weights)
if (type_ == VarType::discrete) {
xx = xx.array().round();
}
loglik_ = (this->pdf(xx, false).array().log()).sum();

if (w.size() == 0) {
w = Eigen::VectorXd::Ones(xx.size());
}

loglik_ = (this->pdf(xx, false).array().log().array() * w.array()).sum();
if (prob0_ > 0) {
// For zero inflated data, all observations with value 0 have been removed,
// so their likelihood contribution is missing. There were n * prob0_ such
// observations, each with log-likelihood contribution log(prob0_).
loglik_ += static_cast<double>(x.size()) * prob0_ * std::log(prob0_);
}

// calculate effective degrees of freedom
interp::InterpolationGrid infl_grid(
grid_points, fitted.col(1).cwiseMin(3.0).cwiseMax(0), 0);
Eigen::VectorXd influences = infl_grid.interpolate(xx).array() * (1 - prob0_);
edf_ = influences.sum() + (prob0_ > 0);
grid_points, fitted.col(1).cwiseMin(3.0).cwiseMax(0), 0);
Eigen::VectorXd influences = infl_grid.interpolate(xx).array();
edf_ = influences.sum() + static_cast<double>(prob0_ > 0);

// store bandwidth in standardized format
bandwidth_ = bandwidth_ / multiplier_;
Expand Down Expand Up @@ -583,7 +595,10 @@ Kde1d::fit_lp(const Eigen::VectorXd& x,
grid_points(m - 1),
m - 1,
Eigen::VectorXd::Ones(x.size()));
wbin = wcount.cwiseQuotient(count);
wbin = (count.array() == 0).select(
Eigen::VectorXd::Zero(count.size()),
wcount.cwiseQuotient(count)
);
}

Eigen::MatrixXd res(f0.size(), 2);
Expand Down
Loading