Fabian Hoppe
Fabian Hoppe
### What happened? QR decomposition fails for sufficiently large matrix/sufficiently small number of processes **if split=1.** For split=0 this behaviour does not occur. Example: The code snipped below produces an...
**Feature functionality** The idea is to introduce a function `ht.unfold` in the manipulations-module that has the same functionality as [PyTorchs unfold](https://pytorch.org/docs/stable/generated/torch.Tensor.unfold.html). In fact, this only requires additional efforts for the...
### Due Diligence - General: - [X] **base branch** must be `main` for new features, latest release branch (e.g. `release/1.3.x`) for bug fixes - [X] **title** of the PR is...
This PR addresses #1243 Description: TBD ... see comments below
This PR is intended to add a triangular solver (see, e.g., `torch.linalg.solve_triangular`) to Heat; see #1096. ## Due Diligence - General: - [X] **base branch** must be `main` for new...
## Description ### Changes in the implementation of QR: * `split=1`: column-block-wise stabilized/modified Gram Schmidt; thus any shape allowed. * `split=0`: TS-QR ("tall skinny QR"); thus input must be tall-skinny...
## Description In this PR we add a class `DXarray` that is somehow similar to `xarray.DataArray` but with `heat.DNDarray`'s instead of `numpy` arrays; whenever possible, we use existing Heat-operations or...
**Feature functionality** Provide a solver for the linear equation $Ax = b$, where $b \in \mathbb{K}^{m \times n}$, $A \in \mathbb{K}^{m \times m}$ are two-dimensional `DNDarrays` such that $A$ is...
### What happened? The function `heat.load` seems to behave quite different w.r.t. scaling depending on whether `split=0` or `split=1`. I tested with a $\approx$ 20GiB data set on different machines...
This PR is intended to contain all changes that are required for introducing a spack module for Heat: in particular we need to mention this in our README and the...