search

jrleeman / rsfmodel

Rate and state frictional solver
MIT License
31 stars 12 forks source link

Critical points #32

Closed jrleeman closed 9 years ago

jrleeman commented 9 years ago

Automatically determines critical times for integrated based upon the acceleration of the load-point. Will handle uneven time spacing.

jrleeman commented 9 years ago

Not sure why Travis is failing. All tests run fine for me locally.

dopplershift commented 9 years ago

Run fine on my mac and on an ubuntu 15.04 64-bit vm. Might want to test against Ubuntu 12.04 64-bit, which is what Travis runs. Have you ever used virtual box?

jrleeman commented 9 years ago

I use parallels. I made a 12.04 machine and followed the way Travis does things and all tests pass... Hmmm

On Jul 13, 2015, at 3:25 PM, Ryan May notifications@github.com wrote:

Run fine on my mac and on an ubuntu 15.04 64-bit vm. Might want to test against Ubuntu 12.04 64-bit, which is what Travis runs. Have you ever used virtual box?

— Reply to this email directly or view it on GitHub.

jrleeman commented 9 years ago

Since this is failing on a clean build 12.04 machine, should be contact the folks at Travis? I've played with it some and can't find any knobs to turn to fix this.

dopplershift commented 9 years ago

On a whim, can you try adding this to the .travis.yml:

addons:
  apt:
    packages:
    - libblas3gf
    - liblapack3gf

This is telling it explicitly to install blas/lapack.

jrleeman commented 9 years ago

Just did (you can see results on Travis page). No dice.

dopplershift commented 9 years ago

Hmmmm. What about:

sudo: required

This will run on a VM rather than a Docker container. Just trying to reduce differences (really changing this should make no difference at all) between what you have working locally and what fails on Travis.

Your best bet contacting them would be to ask for for something specific you could use to debug, like a list of the packages they install. I guess another option is to add the command to list all installed packages from apt-get to .travis.yml. That would allow you to better replicate the Travis environment locally, hopefully reproducing the problem.

jrleeman commented 9 years ago

That didn't help unfortunately. Just commenting out the tcrit part made everything happy again, so at least we know what is causing the issue. I'll see if there are some maxstep params to tweak.

jrleeman commented 9 years ago

Ok. If I increase the mxstep argument, everything runs fine. It's strange that it is necessary for Travis, but not on our systems. The problem with the way I did it above is that we can't let the user set mxstep when calling solve:

model.solve(mxstep=5000)

We get a multiple values for keyword argument error.

dopplershift commented 9 years ago

Add mxstep to odeint_kwargs if it's not there. You could also even check that if the user passes it in, the value is in the right range and could warn if not.

jrleeman commented 9 years ago

Okay, I decided to leave the checking out currently. Is most cases a low mxsteps is fine, but for very stiff cases more steps are needed. 5000 (the default I set) should be enough for just about anything. Finally passes build and tests.