David Turner
David Turner
This PR implements CI testing of brainiak and the example notebooks on Princeton research computing cluster della. I am still trying to work out some issues with failing or hanging...
This is a big PR to remove `typecheck-decorator` dependency and replace with [beartype](https://github.com/beartype/beartype). This involved the following changes: - Replace all non PEP compliant type hints and references to typecheck...
The replaces autograd with PyTorch `torch.func.grad` in GradientOptimization. It also removes references to it in a few other places. There is currently only one test that exercises autograd and this...
I am trying to track down an issue with PEC and fitting a new version of the stability flexibility model. Bryant made a couple of seemingly small modifications to the...
Some small modifications to PEC\Optuna RNG seeding interface based on suggestions from @jvesely and @jdcpni. - Hack was removed that set optuna RandomState object under the hood. - Added ability...
Implement additional support for optimization algorithms from hyperparameter optimization libraries. Potentional options could be optuna, scipy optimize, ray tune, or hyperopt.
Currently, PEC in data fitting mode only supports maximum likelihood estimate (MLE). Ideally, we would be able to do MCMC sampling as well. Some difficulties that we will need to...
What happens when more that two edges connect to a single input port? This came up in our weekly meeting and we didn't know the answer.
The current pipeline for PyTorch to MDF conversion utilizes TorchScript compilation of PyTorch models. The TorchScript graph is created with a call to `torch.jit.script`. The TorchScript graph is then traversed...
Currently, we do not perform any linting of MDF models to ensure they conform to constraints within the specification. We currently accept any model that conforms to MDF's JSON schema...