Carlo Lucibello

saying hi

4

Hi @etotheipluspi @evizero @denizyuret @ilkerkesen @ozanarkancan @sglyon, I've been writing some deep reinforcement examples using julia and knet and I'm pinging you just to let you know, since i think...

- [ ] A3C - [ ] A2C https://blog.openai.com/baselines-acktr-a2c/ - [ ] ACKTR https://blog.openai.com/baselines-acktr-a2c/ - [ ] DQN - [ ] Double DQN https://arxiv.org/abs/1509.06461

graph plotting

1

Introduce plotting through graphviz as Catlab.jl is doing (the code in fact is taken from there). todo: - [ ] add docs - [ ] add tests

It would make sense to have as the default type the more flexible `Network` instead of `Graph`, so that by default one is able to have graph/edge/vertex property maps.

Here is a partial list of missing algorithms. For hard problems (as most problems in this list) both exact and heuristic solvers should be implemented. ### Problems not yet addressed...

It would be nice to have ```julia for (e, val) in emap .... end for (v, val) in vmap .... end ``` An option for an iterator would be ```julia...

edges in collect(edges(g)) should be ordered by index

1

for the Network type, and more generally for graphs with indexed edges. Right now edges are iterated in lexycographic order for (src,dst): ```julia julia> g= Network(10,20) Network(10, 20) with []...

At the present time it creates a Dict. The same holds for eprop(g, name, e).