sandialabs / pyttb / 5647038805

98%

push

github

Gcp opt (#206)

* CP_APR: Package heirarachy
* There shouldn't be a need to access all the cp_apr helpers from top level

* GCP Samplers: Factor out sampler setup from gcp_opt

* FG_EST/Samplers: Fix a few bugs
* Sparse dot product not elementwise multiplication
* Add int type for index
* Fix typing inference

* GCP Optimizers: Add skeleton for gcp optimizer classes
* Will fill in Adam/Adagrad for interface next
* gcp_opt can then more generically call optimizers

* GCP Optimizers: Adam/Adagrad addition
* Still need to find a clear way to test failed step
* Some minor cleanup discovered in testing

* Samplers: Simplify interface
* Masking isn't supported for stochastic solve so just extract zerosnonzeros rom data

* Optimizers: Test Coverage
* Force a failed epoch to exercise reset behavior

* Optimizers: Add LBFGSB Support
* Right now this is the only non-stochastic solver, should generalize if that changes
* Small fix to ktensor for better supported around sorted modes
* Scipy doesn't have type stubs :(

* GCP/Support Fixes:
* Set default sampler in fg_est
* Fix off by one error in our random subs draw
* Resolve row vs column vector disagreement in estimate

* GCP_OPT: Finally add driver

* Documentation: Add autodocs to sphinx for gcp_opt

* Documentation: Update and cleanup gcp_opt documentation.

* Packaging: Fix package discovery now that we have nested directories

327 of 327 new or added lines in 7 files covered. (100.0%)

4445 of 4463 relevant lines covered (99.6%)

2.99 hits per line