Summary
- get some initial code thoughts up.. (details)
- quadrature params stuff (details)
- personal add to gitignore (details)
- merge in branch (details)
- minor (details)
- done plots for BOD (details)
- working on ice model (details)
- minor testing (details)
- working on multi-dimensional ice model (details)
- stopping crit based on first iteration. (details)
- pre_solve and post_solve methods (details)
- working on adaptivity for spatially varying ice model (details)
- push for daniele (details)
- modified interface for refinement (details)
- modified refinement for sparse grid (details)
- initialized map to identity prior to minimize_kl_divergence first call (details)
- move sparse grid adaptation (forming the sparse grid) into 'quadrature' (details)
- current tests on BananPlots (details)
- add comments for myself for tomorrow (details)
- working on ice settin (details)
- 'GradientDimensionAdaptiveSparseGridKLMinimizationValidator' (details)
- working on ice model (details)
- working on whitening of ice model (details)
- working on ice (details)
- better inverse map for ice problem (details)
- minor (details)
- sparse grid is not extracting f from kwargs, to be passed every time (details)
- preparing for run (details)
- trying modification to variance diag calculation (details)
- try except in _refinement of KLAdaptivity (details)
- sign error. BOD (details)
- working on ice (details)
- modified external interface for (x,w) for sparse grid library. (details)
- BananaPlots works perfectly. (details)
- BananaPlots MC vs SG. curently the MC doesnt work because in my branch (details)
- just some printing for diagnosing (details)
- consistent naming (details)
- bod plots for mcqmc (details)
- add the points on the top of the plots for BOD sparse grid (details)
- kwargs (details)
- Is this a bug? (details)
- setting results on paper (details)
- add arguments to pre/post solve (details)
- testing ideas... (details)
- add x,w, to dictionary after error estimation (details)
- validators (details)
- finished with tests for this branch. starting new branch for quadrature (details)
- pushing commented out code that shows the error of log(0) (details)
- changed tolerance for bod problem (details)
- figuring out tolerances (details)
- Remove class GradientDimensionAdaptiveSparseGridKLMinimizationValidator. (details)
- comments, and justfication for changes with comments for Daniele (details)
- remove some unnecesssary code. (details)
- remove more comments, etc (details)
- testing default_initial_quadrature. (details)
- Trying to case on whether or not 'initial' grid was laready created. (details)
- some order in the tolerances (maybe) (details)
- sort weights by absolute value for more stability in order of addition. (details)
- Place plotting under a boolean for toggling ease. (details)
- commenting and refining quadrature... (details)
- clean up quadrature qtype 4 (details)
- remove 'np.newaxis' everywhere. use [x], now that I understand the (details)
- no need for 'previous_fval' under current algorithm.. (details)
- KeyError exception for default quadrature (details)
- KeyError for presolve. (details)
- coment an issue with this branch. why is pull_tar.n_coeffs missing if we (details)
- add ref_params to post_solve (details)
- add ref_params to post_solve, logger info (details)
- Fix BananaPlots for current interface (details)
- Oops, I accidentally merged branches incorrectly. Just rolling back some (details)
- add stochastic volatility to this branch. (details)
- trying much simpler interface to functions/integrands for sparse grid (details)
- bug fix (details)
- space (details)
- comments (details)
- add stochastic volatility to this branch (details)
- change tol for StochasticVolatility.. decrase max_grid_pts to alleviate (details)
- rethinking instantiation of parameters for GaussianDistribution (details)
- reshaping the quadrature function in GaussianDistribution (details)
- using bfgs in adaptivity for now to make sense of tolerances. working on (details)
- we don't need 'converged', or 'termination reason' or the integral (details)
- BFGS for the BOD problem (details)
- now adaptive_quadratures can be built also wrt for pushforward/pullback (details)
- relegate checking for uniqueness if x,w to some 'debug' mode in future (details)
- start testing laplace approximation. new BOD file for that. (details)
- added missing file (details)
- mirroring what we've done in frozendistributions quadature, dont need to (details)
- adapter -> qparams['adapter'] (details)
- techincally dont need two files, just separate them though so we can (details)
- nax = np.newaxis (details)
- standard normal doesn't need trnasformation (dont need to use sampling (details)
- right now quadratur requires abs_tol, even if initial grid is just an (details)
- pre/post solve fixes try except - > if else (details)
- pre/post solve fixes (details)
- Decrease optimizaiton tol for the Laplace Approx problem, since its (details)
- testing parameters for initial_quadrature (details)
- I believe you need to warmstart here. (details)
- use error estimation points in post_solve (details)
- allow first variation/line serach to use the last set of points from (details)
- add L^2 regularization part of objective gradient. this right now is (details)
- add regularization to the laplace approx example (details)
- Revert "add L^2 regularization part of objective gradient. this right (details)
- slash \, and justification for qparams abs_tol (details)
- add norm printing (details)
- more informative comment. realized that the regularization is (details)
- stuff for l2 regularization (details)
- parameter adjustments: decrease regularizaztion, relax the initial (details)
- make the optimization a bit easier for this regularized problem. too (details)
- setting up Poisson (details)
- almost done with the constant field Poisson setting (details)
- fixing gradients (details)
- Poisson problem set up for constant field and Gaussian field (details)
- whitening script for Poisson (details)
- added action of the Hessian (details)
- fixed whitening script and added xml map descriptor (details)
- fixed bug in FrozenDistributions qtype==0 (details)
- change how w outputs in darkspark (details)
- increase max allowed levels... all quadarture levels were maxing out at (details)
- increase max allowed levels to a very high amount.. (details)
- fix a bug (details)
- working on sv (details)
- working on new active variable detection technique (details)
- turn try except into if/else. comment out new metho for prunning of (details)
- abs_tol is not needed for the default quad case. bring back 'self.eps', (details)
- if BFGS with inf norm, print inf norm of jacobina, not 2 norm. (details)
- eps in validator (details)
- working on SV. Try ½ solve tol to start with. ½ each time. set a minimum (details)
- prints (details)
- Informative printing of info, revert to 1/10 solve_tol, seems to work. (details)
- start MC SV (details)
- working on sv tests for talk (details)
- figures generated for Josh talk (details)
- update poisson example for newer versions of fenics (details)
- undo bug (details)
- start with diagonal (details)
- Fixed Poisson problem set up in SG adaptivity (details)
- made DARKSparK and optional External package (details)
- mpi works on the poisson example (details)
- cleaning up poisson problem to be included to the available (details)
- working on general setting of PDE problems with FENICS (details)
- going on... (details)
- going on.. (details)
- going on.. (details)
- almost done with the Poisson setting (details)
- added LinearTriangularMap for fast regression (nested Cholesky) (details)
- working on setting for Poisson problem (details)
- working on Poisson (details)
- added the bootstrap validator and changed arguments on error_estimation (details)
- added bootstrap option to tmap-tm (details)
- working on the bootstrap validator (details)
- changed the stopping criterion for the bootstrap validator (details)
- polishing adaptivity (details)
- fixing coefficients update and cache usage in first variation (details)
- added better estimation of points multiplier for bootstrap validator (details)
- working on randomized approximation of Hessian for Laplace approximation (details)
- randomized_direct_svd works (details)
- randomized svd is consistent with eigenvalue decomposition (details)
- low-rank laplace work with outer gradients and low-rank hessian (details)
- implementing likelihood informed subspace procedure (details)
- working on Laplace approximation with lis in the head of the square root (details)
- added option lis on tmap-laplace. LinearTransportMaps and (details)