Mean marginal likelihood
Important note: This package dependends on the package fastkde, which must have been installed previously. It also needs densratio_py if you want to use LSCDE (otherwise, just delete it from the package directory and init file).
The purpose of this package is to provide tools for computing trajectory acceptability scores based on the Mean Marginal Likelihood. More information on this criteria and its uses can be found in the following papers: [1], [2]. This package contains classes and methods allowing to * compute marginal densities from a set of trajectories, * plot the corresponding heatmap, * compute the overall mean marginal likelihood scores of test trajectories using normalized densities scaling and confidence level scaling, * plot the marginal likelihoods point-by-point for a list of curves, ...
While the module fastkde_mml computes the MML based on a nonparametric density estimator called the Self-consistent kernel estimator (see [3]), Gaussian Mixtures Models (whose number of components is set between 1 and 5 using BIC) are used in the module gmm_mml. The later allows to export the parameters of each GMM as a csv file.
Modules LSCDE_mml and FPCA_mml contain code capable of computing trajectories likelihood based on Least-Squares Conditional Density Estimation and Functional PCA (see comparison results in this article [1]).
Pre-requisites
- Python 2.7
- sklearn
- pandas
- seaborn
- joblib
- Numpy
- fastkde
- densratio_py
Overview
A tutorial notebook can be found in this same directory, under the name Tutorial_traj_accept.ipynb.