Max Balandat
Max Balandat
There is some confusion among some users regarding the distinction and usage of constraints (see e.g. #432). We should write up a doc section explaining them: How are parameter constraints...
# 🚀 Feature Request We'd like to support a simplex constraint on a subset of parameters (all parameters positive, sum to 1). ## Motivation See e.g. discussion in #480
Using `fit_gpytorch_model` on a model obtained using `condition_on_observations` produces bogus fits. This was described in https://github.com/pytorch/botorch/issues/337#issuecomment-563358045. Basic model: ``` import torch from torch import Tensor from botorch.models import SingleTaskGP from...
`test_swa` fails with the following: ``` test/test_swa.py::TestSWA::test_swa /Users/balandat/Code/contrib/torchcontrib/optim/swa.py:200: UserWarning: SWA wasn't applied to param tensor([[-1.3171e-01, 3.3898e-01, 3.9688e-01, 9.4115e-01, -1.0891e-01], [-6.1716e-01, -9.9478e-02, -3.5716e-01, 6.9384e-01, -2.1573e-01], [-9.2800e-01, 2.5287e-01, -9.5098e-01, -3.5055e-01, 6.5104e-01], [...
running `test_swa.py` on a device with multiple GPUs results in the following: ``` Test output: > test_swa (test.test_swa.TestSWA) ... ERROR > > ====================================================================== > ERROR: test_swa (test.test_swa.TestSWA) > ---------------------------------------------------------------------- >...
It would be nice to have a function, say "N( )", that indicates that input data with pandas dtype "category" should be interpreted as numerical data. In that sense this...
### Issue Description When running NUTS on a pretty basic model using `1.12.1+cu113`, the following exception is raised (for a full repro see the colab link below): ``` --------------------------------------------------------------------------- RuntimeError...
Summary: Unfortunately, scipy does not allow timing out the optimization based on wall time. This diff implements a lightweight wrapper around `scipy.optimize.minimize` to achieve this. The new `minimize_with_timeout` method calls...
BoTorch's `FixedNoiseGP` has been deprecated in favor of a more general `SingleTaskGP` but still appears in Ax, e.g. in the modular Botorch model tutorial: https://ax.dev/tutorials/modular_botax.html Let's remove it there and...
Using `fixed_features` in [`Modelbridge.gen()`](https://github.com/facebook/Ax/blob/main/ax/modelbridge/base.py#L748) allows one to fix the values of certain parameters. This is useful in certain "contextual" optimization setups, see e.g. #1931 for a discussion. Currently, this is...