surgroup / uqpy Goto Github PK

• Make the default to write minimal text, but allow functions to output some information text if requested.

• Move the 'Diagnostics' to the Utilities module

• Document and push to Development

• Add Documentation to Users Guide

• Modify setup.py to include matplotlib

Thank you for your effort and for making the code publicly available! I'm wondering if you have an estimate when v2.1.0 of UQpy will be released?

• Modify refinement criteria to incorporate uncertainty in Jacobian

• Code GE-RSS with Voronoi cells

• Add error checks

• Doc string

• Push to the development branch

• Modify user manual

[ ] At this point, I thought MCS only sampled independent RVs. But, there is now an option to input a copula. Does this work?

• There is a comment "# ne need to do other checks as they will be done within Distributions.py" Should read "No need..."?

• Modify the input to RunModel to allow fixed parameters to be passed in.

• Add Importance Sampling class to SampleMethods

• Create documentation for the class

@bsaakash

• Convert .tex to .rst if possible
• Host the documentation on a separate branch
• Resolve issues of building the documentation

• Figure out how this got broken and fix the issue.

• Modify the RunModel module itself to interface with a single Python script

• Build template Python script

• Matlab model currently in the examples

• Abaqus model

• Users Guide needs to be updated to reflect the changes

• Pushed to the master

• Resolve differences in Maximum Likelihood estimation

• Place it in the appropriate module (Inference or Utilities?)

• Document and submit to Development

• Update Users Guide.

• RSS - Mohit

• Inference Class - Audrey

• Subset Simulation - Mike

• Taylor Series - Dimitris

• Include gradient of the objective function to estimate hyperparameters

• Add error checks

• Doc string

• Finalize four additional correlation models

• Push to the development branch

• Modify user manual

• Resolve the collapsing dimension of numpy arrays throughout the code so that (nsim,) does not arise

• Extract all code from the development branch that relates to terminal processing.
• Consolidate all of this code into a single module on Dev_Dimitris
• From this, build UQpy_Parser module for V1.1 or V1.2

@dgiovanis

• Write in CoV in MMH for subset simulation

• MCS
• LSS
• STS

• Error

• MCS

• RunModel

• Distributions

• LHS

• Correlate

• Decorrelate

• Nataf

• InvNataf

• SRM

• BSRM

• KLE

• Translation

• InverseTranslation

Each section of the Users Manual should have a small discussion of theory behind the method. This should be just enough to understand the inputs and outputs, without derivations (typically no more than one page, see e.g. Nataf). If the theory is complex and requires elaborate discussion, references should be provided that give adequate background to the user.

• MCS

• LHS

• STS

• MCMC

• Correlate

• Decorrelate

• Nataf

• InvNataf

• SROM

• Kriging

• SubsetSimulation

• TaylorSeries

• SRM

• BSRM

• KLE

• Translation

• InverseTranslation

• BayesModelSelection

• InfoModelSelection

• BayesParameterEstimation

• Update SROM examples to include the eigenvalue problem with RunModel

• Push to development

• Modify Users Manual to reflect the new example

• Add a short "Theory" section

• Update the front page material on the UQpy GitHub site.

To be consistent with existing literature

• Complete the TaylorSeries for FORM/SORM analyses.

• Complete documentation for TaylorSeries

@zhangjiaxin2012

• Full implementation of Bayesian and information-theoretic multimodel inference
• Optimal importance sampling estimator and importance sampling in SampleMethods
• Documentation and comments
• Full documentation in UsersGuide

• Add LSS/PSS/LPSS classes to SampleMethods

• Allow the user to pass a seed value for Subset Simulation with MMH or MH

• Update the introductory sections of the UsersGuide including installation, methods of operation, etc.
• Installation section
• Adding functionality to UQpy
• Structure of the code
• Modes of operation

• Include gradient of the objective function to estimate hyperparameters

• Add error checks

• Doc string

• Push to the development branch

• Modify user manual

• Figure out exactly how to do autotesting for each class.

• How much extra work is it to implement this in our development?

• Add coefficient of variation calculation to the affine invariant version of Subset Simulation

• MPI interface for communication between nodes. Do we need to use that?

• SLURM - Generate SLURM compatible files directly

• Pass kriging as an object.

• Make self.lf a function itself. That way, it can be passed in or it can be assigned to one of the internal functions based on the keyword lf.

• Allow the code to add more than one point in a give iteration.

• Rewrite the code so that each learning function is called directly as a function through self.lf.

• Complete the jupyter script examples.

• 'tools' is a redundant module and needs to be removed. Anything that is not already moved, should be moved to 'Utilities'

@mohitcek

• Make sure SROM is fully consistent with other classes (e.g. MCMC) and modules (RunModel)
• Fully comment the code (both for documentation and throughout the code)
• Build documentation in UsersGuide
• Build representative Jupyter script examples

I can successfully create and use a Kriging surrogate model in UQPy, but I can't save it to a pickle file. When I try, I get the following error message:

Traceback (most recent call last):
  File "./mk_gp_uqpy_fail_to_pickle.py", line 17, in <module>
    pickle.dump(gp, pkl_file)
AttributeError: Can't pickle local object 'Krig.init_krig.<locals>.regress.<locals>.r'

A workaround for this is to use the third-party module dill instead of the standard module pickle, but it would be nice to not need to do that.

@lohitv96

• - Make sure LHS is fully consistent with other classes (e.g. MCMC) and modules (RunModel)
• - Fully comment the code (both for documentation and throughout the code)
• - Build documentation in UsersGuide
• - Build representative Jupyter script examples.

(FYI, this is a follow-up to a problem I had e-mailed mds2120 about directly.)

In the method run_lhs() defined in UQpy/SampleMethods.py, line 203 reads as follows:

 samples_u_to_x[:, j] = i_cdf(samples[:, j], self.distribution[j].params)

However, this is wrong, and running run_lhs() leads to an exception with the error message: AttributeError: 'Distribution' object has no attribute 'params'.

The fix for this is straightforward. Change the line to read as follows:

 samples_u_to_x[:, j] = i_cdf(samples[:, j], self.dist_params[j])

Then, run_lhs() should work as intended.

• Remove "dimension" from input attributes

• Recall the reason we need to explicitly state dimension - was it only because of ambiguity in reading text file.

• Resolve this so we don't need to explicitly state dimension

• Put error checks into classes requiring matrix reshaping.

• Resolve the evidence calculation

• Build BayesModelSelection class in Inference

• Document code and push to Development

• Update Users Guide

• Figure out how to host the examples on Github or elsewhere

Push Subset simulation to development branch

• Figure out how to put the code on PYPI and install using pip.

• Better evidence calculation methods in Inference Module

• Externalize evidence into its own class.

• Will require modifications to MCMC class.

• Build pdfs, cdfs, inverse cdfs, etc. as needed for the code
• Change pdf_type, pdf_params, etc. to dist_type, dist_params, etc throughout the code

• Structure distributions by treating each family using SubDistributions

• Create product distributions using SubDistributions

• Create joint distributions from copula

• Push to development

• [ ]

• When changes were made to MCMC, the documentation was never updated to reflect those changes.