Mike J Innes
Mike J Innes
I'd like to use this excellent package for the filters commonly used in deep learning. I tried a naive implementation [here](https://github.com/FluxML/NNlib.jl/pull/9/files), but performance isn't great and I'm wondering if we...
The `set_tokenizer` API seems a bit suspect here, given that it can be replaced with ```julia const tokenize = WordTokenizers.nltk_tokenize ``` and likewise for RevTok etc, without bringing in multiple...
In general this proposal looks great. However, I don't understand why traps can't be caught. The proposal states: > The rationale for this is that in general traps are not...
After #190, `install` was broken due to some missing imports. Also, the behaviour of `unpack` seems to have changed: it now errors when used this way, because the target directory...
(as we are doing in the rest of the JuliaGPU ecosystem).
This is a bit more reliable than the current build; on at least one system CUDNN.jl couldn't find libcudnn where CuArrays finds it fine. (Aside from that it's just nicer...
Although it's great to have ML-specific functionality held separately from base CuArrays in this package, I'm wondering if we should keep convolutions in CuArrays itself (effectively making it part of...
Continuing our series "cool things we can't have yet", and inspired by [this comment](https://github.com/FluxML/Flux.jl/pull/607#issuecomment-461889413) I was thinking about how we'll expose BPTT. Currently, given a forward pass like this: ```julia...
I'm struggling to download the trained models from the Baidu page linked [here](https://github.com/zhaw/neural_style/tree/master/perceptual) – it gives me a very slow or 0 download speed when I try it. Would be...
by passing an io object around, rather than opening the file many times. Also, in this case I've read the whole file at once for maximum read speed. I've tried...