JuliaReach/Lobby

Calculating Reachable states of Dynamical Systems in Julia

Marcelo Forets

@mforets

we regularly skype with the rest of the team (schiliic, dpsanders, lbenet), so i'll put this topic in the "agenda" :)

@ChrisRackauckas and also @schillic : this week with sebastian we had a couple of sprints to understand and model a hybrid system from https://discourse.julialang.org/t/differentialequations-error-event-repeated-at-the-same-time/51830 using propagation techniques. not sure we'll finish this week because of family time. but it was much fun as it involves modeling nonlinear (periodic) guards, and we can exploit sparsity of the constraints to make an efficient solver

Christian Schilling

@schillic

:+1:

Christopher Rackauckas

@ChrisRackauckas

Ahh that's cool

I haven't looked into that system yet

Marcelo Forets

@mforets

FYI the PR JuliaReach/ReachabilityAnalysis.jl#390 contained several unrelated changes so i decided to split into smaller ones. i've started to cleanup the codebase with respect to changes that are recently available in LazySets

also the RA internal methods for disjointness, intersection and inclusion are messy so i'll fix that too

Marcelo Forets

@mforets

i ran into a nice application of the @concretize stuff: https://github.com/JuliaReach/ReachabilityAnalysis.jl/pull/397/files#diff-092a615958f0d53a3518d6754e427a61a7228dc6e0c086d53a7a3ef562b314a4R300

Marcelo Forets

@mforets

a new article on verifying LTI systems neural network controllers https://arxiv.org/pdf/2101.01815.pdf

Marcelo Forets

@mforets

hi, this week i'm back to package development. i was planning to address the Star sets implementation this afternoon. i've got like 2 hours :smile:

Christian Schilling

@schillic

good luck :)

Marcelo Forets

@mforets

there is a nice boost on the startup time:

[mforets@localhost bin]$ julia
               _
   _       _ _(_)_     |  Documentation: https://docs.julialang.org
  (_)     | (_) (_)    |
   _ _   _| |_  __ _   |  Type "?" for help, "]?" for Pkg help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 1.5.2 (2020-09-23)
 _/ |\__'_|_|_|\__'_|  |  Official https://julialang.org/ release
|__/                   |

julia> @time using LazySets
  5.681156 seconds (11.94 M allocations: 694.502 MiB, 4.23% gc time)

julia> exit()
j[mforets@localhost bin]$ ./julia
               _
   _       _ _(_)_     |  Documentation: https://docs.julialang.org
  (_)     | (_) (_)    |
   _ _   _| |_  __ _   |  Type "?" for help, "]?" for Pkg help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 1.6.0-beta1 (2021-01-08)
 _/ |\__'_|_|_|\__'_|  |  Official https://julialang.org/ release
|__/                   |

julia> @time using LazySets
  2.215739 seconds (5.24 M allocations: 375.817 MiB, 1.47% gc time, 0.22% compilation time)

Marcelo Forets

@mforets

i merged some LazySets PRs without waiting for the travis CI to finish . this is against our rules i know ... but we can't wait 40 mins - 1hour for each single PR it is just too slow. maybe if we change the CI to github actions it reduces

in RA it reduced the waiting time by half

or we get a university machine to run the tests and upload the results in less than 10 minutes

is that possible? for the machine i can ask @dfcaporale

Marcelo Forets

@mforets

it would be limited to only 1 platform though, but that is less of a problem (we don't have system-specific code, but some of our optional dependencies do)

Christian Schilling

@schillic

for me waiting one hour is not a blocker because 1. most often the branches are independent and if not then 2. i can merge the branches into other branches locally, so i don't have to wait for CI to finish to continue developing. but i don't want to impose that

about github actions: would be great :)

Marcelo Forets

@mforets

you're absolutely right. sometimes it feels like having 3 approved PRs waiting the CI to finish (and you know it's OK 99%) is too much

there's satisfaction if you can close issues before you call it a day :laughing:

Christian Schilling

@schillic

true :)

Marcelo Forets

@mforets

but, yeah anyway there is plenty of advantage of having these online tools, for example that they provide hyperlinks to each line of the build

this tends to be very useful when there are issues

Christian Schilling

@schillic

how much work is it to try github actions?

Marcelo Forets

@mforets

i would be happy if we are able to setup a github action locally (whatever that means) and have a bot that comments that the tests passed or not, maybe with a log but that may be harder. i'm unfamiliar to this stuff but maybe there are examples that we can replicate

Christian Schilling

@schillic

but you said you did it in RA?

Marcelo Forets

@mforets

yes it is done

i thought you meant something local

for the github actions it is easy

Christian Schilling

@schillic

no no

is there a reason why we don't have it in LazySets?

Marcelo Forets

@mforets

it should be

1) add a script similar to https://github.com/JuliaReach/ReachabilityAnalysis.jl/blob/master/.github/workflows/ci.yml

2) delete .travis.yml

that was it iirc

by the way there is https://docs.github.com/en/actions/hosting-your-own-runners

Christian Schilling

@schillic

maybe let's first try github actions :)

Marcelo Forets

@mforets

yes

we have in https://github.com/JuliaReach/LazySets.jl/blob/master/.travis.yml#L31

that build_private script

i dont know how to implement it in the github action script

Christian Schilling

@schillic

ah :(

Marcelo Forets

@mforets

what it does is https://github.com/JuliaReach/LazySets.jl/blob/master/deps/build_private.jl#L6

we will remove it once we pass to v1.6 only

Christian Schilling

@schillic

true

Marcelo Forets

@mforets

we can wait until then :)

Where communities thrive

People

Repo info

Activity