-
While working with LSTM I noticed that Recur is not type stable for 3 dimensional arrays.
If we execute:
```jl
using Flux
model = LSTM(1=>1)
x = rand(Float32, 1, 1, 1)
@code_warntype model(x)
…
-
Have you implemented `Base.repeat` for CuArrays? `repeat` seems to use the slow fallback. Is it hard to? Thanks!
-
## Description
When accessing the adjoints for `var` types, the same member function is used for both scalars and containers:
```cpp
double adjoint = var.adj();
Eigen::VectorXd adjoint_vec = var…
-
We can add the Cholesky parameterization on both X and Sigma or just on Sigma
In cpp we'd use LDLt so we don't have to `add_diag`.
```stan
real wishart_cholesky_a_lpdf(matrix X, real n, matrix …
-
| | |
|------------------|-----------------|
|Previous ID | SR-9345 |
|Radar | None |
|Original Reporter | jekbradbury (…
-
When taking the jacobian of a sample FFT-Pipeline, `Zygote` complains about missing fields in the corresponding FFT Plan:
```julia
using Zygote, FFTW
n = rand(100);
f = plan_rfft(n);
fi = pla…
-
Are there any known issues that could prevent a switch? If not, maybe consider switching the default AD to ReverseDiff after https://github.com/TuringLang/DistributionsAD.jl/pull/44 is merged? Based o…
yebai updated
2 years ago
-
As a first step leading up to GRU-ODE or ODE-LSTM implementations, I'd like to switch out the Dense layer in the neural_ode_sciml example with a GRU layer. However, doing so raises the error `LoadErro…
-
**Describe the bug**
Adding a sparse matrix adjoint to a dense or sparse matrix results in kernel errors.
**To reproduce**
The Minimal Working Example (MWE) for this bug:
```julia
A = c…
-
Hi, I'm trying to install idealstan on R 4.0.0 64-bit Windows. I'm getting an error when trying to install it, though.
```
`> devtools::install_github('saudiwin/idealstan',local=F)
Downloading GitH…