Kyle Daruwalla
Kyle Daruwalla
Currently, it seems like the only way to use PublishThemes is to override the following before building the docs: ```julia Publish.Themes.default() = artifact"my-theme" ``` This also applies to any theme...
In FastAI.jl, our examples/tutorials require training on a GPU. Running this code locally in a Jupyter notebook, then committing the notebook w/ output to the repo seems like the best...
None of the math blocks are rendered using KaTeX in my deployed page (see below). Is the JS loaded by default, or is there something required on my part? I...
This is a WIP in response to https://github.com/JuliaML/LearnBase.jl/pull/44. The changes to the LearnBase.jl interface reduce the MLDataPattern.jl codebase significantly. Most notably, we are able to avoid a lot of the...
In Flux, we have `trainable` to designate a subset of leaves as nodes to walk when updating parameters for training. In FluxPrune.jl, I defined `pruneable` to designate a subset of...
**Is your feature request related to a problem? Please describe.** Currently, I use sessions to save "snapshots" of my browser when I was working on a specific project. When I...
This adds a "metadata" `PropDict` to `Learner` for storing information that is required for training but extraneous to the training state or callback state. This is useful for unconventional training...
`LogHyperparams` relies on `:hyperparams` being defined as a key in `learner.cbstate`, but this is only initialized by `Scheduler`. So, trying to use a hyper parameter logging callback without scheduling callback...
I am not really sure what is going on, since all the other logged values look fine. I have a simple supervised training flow with the standard phases. Here is...
Using this like: ```julia function mylayer(x) rng = rng_from_array(x) ... end ``` will try to propagate gradients through which we don't want.