pshashk

Thanks for the offer. Unfortunately, right now I don't have enough spare time, but I will be glad to collaborate in the future.

On the producer side, it seems like a simple state check is enough. ```julia create_connection() = connection(;virtualhost="/", host=HOST, port=PORT, auth_params=AUTH) create_channel(conn) = channel(conn, AMQPClient.UNUSED_CHANNEL, true) mutable struct Producer conn chnl...

I've managed to implement consumer that could handle broker restarts. But I doubt that's the best way to do it. ```julia function consume() while true try conn = connection() chnl...

[jug](http://bart6114.github.io/jug/index.html)? There is also a node.js based [plugin](https://github.com/Bart6114/jug.parallel) for parallel processing, but I haven't tried it. Things like request [memoization](https://github.com/r-lib/memoise) and approximate nearest neighbor search (e.g., [Annoy](https://github.com/eddelbuettel/rcppannoy) if L2 norms...

Somehow `*Triangular` gradient definitions overwrite each other and `UpperTriangular(::TrackedArray)` returns tracked `LowerTriangular`. Minimal example with current master: ```julia using LinearAlgebra, Tracker using Tracker: track, data, @grad using LinearAlgebra: UpperTriangular, LowerTriangular...

I think, for performance and ease-of-use reasons, it is desirable to have a container type for handling batches of variable-length sequences similar to pytorch PackedSequence or tensorflow RaggedTensor. Otherwise, the...

> @pshashk Other than from bringing more high level API usability, are you seeing benefits from a specific RNN container vs adding a seq_length element to the usual data iterator?...