hippyflow
hippyflow copied to clipboard
hippylib
Dimension reduced surrogate construction for parametric PDE maps
In this PR we added 1. `dataGenerator` module that handles more complicated data generation routines that are needed for function-to-function maps. 2. Data driven computation of active subspace, and POD,...
In the `hf.DataGenerator` class the input basis should be used and the output projector should be used (including Riesz representer). The naming conventions in POD and KLE and AS should...
In the case that the full Jacobian is requested, and it is either short-fat or tall-skinny randomized SVD is inefficient, and its better to compute action on identity. This case...
This is a minor addition, but adding methods that get dense mass and stiffness matrices from the `POD` object would be very useful.
* What needs to be done to ensure the decompositions and numpy representations of multivectors make sense for mesh partitioned problems? So far instance parallelism works, but there could be...
The existing error tests assume that the projectors are identity-orthogonal.
@tomoleary, it'd be nice to be able to pip install `hippyflow` instead of having to create an environment variable to find the library. 1. How much can we copy from...
In the two-step generation the output encoder `MPhi` was not saved properly.
Some bug fixes to the `DataGenerator` - Changed locations in `two_step_generate` where `compress` and `clean_up` flags aren't being used. In particular, `compress = True` for the derivatives are causing memory...
