Tiem van der Deure
Tiem van der Deure
LKJCholesky causes type stability problems. MWE ```julia using Turing, Random @model function simple_model() F ~ LKJCholesky(2, 3.0) end model = simple_model() @code_warntype model.f( model, Turing.VarInfo(model), Turing.SamplingContext( Random.default_rng(), Turing.SampleFromPrior(), Turing.DefaultContext() ),...
I bumped into some more shenanigans with `reversediff` and covariance matrices. Generating a vector of numbers using `arraydist`, multiplying that with a matrix from LKJCholesky, and passing it back to...
Sampling from `LKJ ` or `LKJCholesky` and assigning to a vector of matrices errors, regardless of AD backend. The code compiles but errors with a `BoundsError`. Sampling from the Prior...
When fitting models with a large number of variables, Lasso.jl and GLMnet return different paths, and the difference grows bigger as the number of variables is bigger. An example to...
I need to run GPU operations inside a loop, where the output of one iteration is used in the next one. However, even very simple GPU broadcasts result in a...
This PR adds a separate constructor `NeuralNetworkBinaryClassifier` for binary classification. Unlike for the existing `NeuralNetworkClassifier` constructor, in the final layer always has size 1. By default, a sigmoid function is...
An attempt to avoid the worst floating point errors when calculating step. @rafaqz let me know what you think about this heuristic
Writing a `Raster` to a netcdf and reading it back in again yields a `Raster` with `Explicit Intervals`, even if the raster originally had `Regular Intervals`. E.g. ```julia using Rasters,...
A `Raster` created by concatenating multiple other `Raster`s keeps the `metadata` of the first `Raster`. Is that on purpose? MWE: ```julia using Rasters, RasterDataSources months = [4,5,6] rasters = map(m...
I used to be able to plot RGBA rasters with Makie using heatmap, but some recent version apparently broke this. I now get an error message that says ``` Makie...