Tags: eford/ABC.jl
Tags
Hsu et al 2018 (#2) * Convert to nearest positive definite matrix added - Added Mike Croucher's Python implementation of the Alternating Projections method to finding the nearest positive definite correlation matrix. (Note: Need to move nearest_correlation.py to Julia working directory for PyCall to find it.) - Edited util.jl so that deprecation warnings wouldn't show up. - To Do: Still cannot generate valid thetas within max iterations -> look into moving normalization of radius-period probabilities into update_sim_param_from_vector. * Added try-catch for nearest pos. def. matrix - Serial version of updating population will now only convert covariance matrix to nearest positive definite if the error occurs when calling generate_theta. * Changed to Diagonal Covariance Matrix - Added new class for diagonal covariance matrix - Adjusted alg.jl to use diagonal covariance - Removed try-catch for positive definite * Removed redundant print statement - Removed print statement for original pop weights because it is equivalent to the new pop weights from the previous generation. * Added printing # of planets for particles - Removed printing of population weights * Commented print statement for # planets * Removed usage of PyCall for nearest_correlation * Added Reference to CovarDiagonal * Changed comment for multi-bin cases * Added check for non-improving epsilon * Testing zero detection catalogs * Writes history of epsilon/mean/std dev. in arrays * Changes to allow using Uniform prior - Now checks for array output to is_valid and normalize and for scaler output from pdf. * Change criteria for stopping at plateau - Now checks for 3 consecutive generations without change in epsilon instead of counting 5 total repetitions. - Temporary disabled summary statistics storage. * Correct Counting Up of Repeating Eps * Fix for deprecation warning * testing modifications on green * Merged with upstream 10-11-17 * Deleted temp files created by editor * Misc. edits to alg.jl * Experiments with make_proposal * Changed to 20% median # of attempts for stopping criteria * Final version used for publication. * Removed Python make positive definite code