Problem with spinnaker and local install of nrp


#1

Hi All,

I installed the neurorobotics platform locally (i.e. via docker) which works fine when running the holodeck husky braitenberg experiment on nest. However, I am having problems trying to run the spinnaker version of this experiment. I installed spinnaker on the backend container using the instructions on https://developer.humanbrainproject.eu/docs/projects/HBP%20Neurorobotics%20Platform/2.0/nrp/developer_manual/CLE/spinnaker.html. However, when I try running the holodeck husky braitenberg experiment on spinnaker or the nest version, i now get the following error : “No server can handle your simulation at the moment. Please try again later”.

Can anyone help?

Thanks and best wishes

Paul


#2

Hi Paul,

we are currently working on the NRP Spinnaker Integration for the docker container. Meanwhile, you can get it running by patching the docker container following these steps (this was originally answered by Björn):

./nrp_installer.sh connect_backend
set MPLBACKEND=agg
source /home/bbpnrsoa/.opt/platform_venv/bin/activate
pip install matplotlib==1.5.1
pip install spynnaker8 pyNN==0.9.4
python -m spynnaker8.setup_pynn
python -c "import pyNN.spiNNaker as sim\nsim.setup()\nsim.end()"

# set machineName and version according to your setup
# spinnakermanchester.github.io/spynnaker/5.0.0/PyNNOnSpinnakerInstall.html#-configuration
nano /home/bbpnrsoa/.spynnaker.cfg

cd Experiments
git checkout development

cd ../CLE
git config remote.origin.fetch "+refs/heads/development*:refs/remotes/origin/development*"
git fetch
git checkout development

exit
./nrp_installer.sh restart_backend

# then remove the old spinnaker-experiment (if there is one) and clone it again from the template
# the robot should start moving about 30s after starting the experiment

Maybe you should also do ./nrp_installer.sh reset_backend as a first step.

Best,
Daniel


#3

Hi Daniel,

Thanks for your help.

I have successfully installed spinnaker.

The holodeck husky braitenberg experiment seems to be working on spinnaker now, but the husky robot is moving very slowly, i.e. the robot has turned about 90 degrees in 6 minutes. Is this normal or do I need to adjust a setting somewhere?

Thanks for your help.

Paul


#4

I can confirm, that this happens with happens with the docker-installation.

further testing shows, that this only happens with the spinnaker-experiment, but not with the regular non-spinnaker experiment. Also this doesn’t seem to happen with the non-docker installation of nrp. A significant difference between the docker and the regular installation of NRP, is that in the docker-installation gzserver only does software-rendering, while the regular installation also can render on the GPU.

the CPU-usage in the docker container for the spinnaker-experiment and for the regular experiment looks about the same. (the majority of cpu-load comes from gzserver)

I’m currently installing the development-setup into the docker-container to see if the problem persists in newer versions. (The odds of this working at are slim, but if it does it would help a lot).

I will tell you when I know more


#5

Hi bjorn,

Ok, thank you for your email.

With regards to the non-docker installation of nrp, I am also having problems using this with spinnaker.
I have installed nrp using your source install instructions and installed the spinnaker software following the instructions on https://developer.humanbrainproject.eu/docs/projects/HBP%20Neurorobotics%20Platform/2.0/nrp/developer_manual/CLE/spinnaker.html. I have tried installing both a few times but I get the same problem when i try running the Holodeck Husky Braitenberg experiment on SpiNNaker: all of the following transfer functions are switched off and I cannot switch them on:

csv_spike_monitor
all_neurons_spike_monitor
linear_twist
eye_sensor_transmit

When i try to switch them on I get Runtime brain error messages. For example, when I try switching on the eye_sensor_transit function I get error:

eye_sensor_transmit: “‘module’ object has no attribute ‘red_left_eye’” (TF Loading Exception)

and when I try switching on the linear_twist function I get error:

‘module’ object has no attribute ‘left_wheel_motor’

If you could help with this that would also be great.

Thanks and best wishes

Paul


#6

Same problem here: full spynnaker support is not yet in the release version of NRP. You can either wait for the next release or try to install the relevant parts of the relevant parts of NRP directly from the source code:

cd $HBP/Experiments
git checkout development

cd $HBP/CLE
git config remote.origin.fetch "+refs/heads/development*:refs/remotes/origin/development*"
git fetch
git checkout development

I have not tested this script myself, since I don’t have a matching setup available. Since the same thing works in the docker installation, I’m confident, that it will work in this case, too.

Also note, that this link is outdated (I think). The recommended way to install spynnaker is to use the pypi-packet:

source $NRP_VIRTUAL_ENV/bin/activate
pip install matplotlib==1.5.1
pip install spynnaker8 pyNN==0.9.4
python -m spynnaker8.setup_pynn

(If it works with the old way, there is no need to change it for now)

note: with the next release of NRP none of these steps should be needed and spynnaker should work out-of-the-box (assuming you have a valid ~/.spinnaker.cfg)


#7

Hi Bjorn,

Thank you for your email.

I have tried the code you mentioned above but it fails on the second line, i.e. on:

git checkout development

I get the following error message:

error: pathspec ‘development’ did not match any file(s) known to git.

Is there a solution to this?

Thanks for your help.

Best Wishes

Paul


#8

there will be a new NRP release soon (nextweek, maybe even this week). In this release there will be much better spynnaker support.

if you don’t want to wait, you can try this (I don’t recommend it though)

cd $HBP/Experiments
git config remote.origin.fetch "+refs/heads/development*:refs/remotes/origin/development*"
git fetch
git checkout development

(the same as in the CLE folder)
Please note: Doing these steps are may break your setup (actually the chance for this is quite high because of a mismatch between the versions of CLE and all other subprojects of NPR). And even if it works, you would be working with an unstable version of NRP, because the dev branch is not supposed to be stable.


#9

Hi Bjorn,

Thank you for your email.

Ok, I will wait for the next release. If you could let me know once it’s released that would be great.

Thanks for your help.

Best wishes

Paul


#10

Dear Paul,

A new version (2.3.2) has been released for docker installation. You can try it by updating with:

./nrp_installer update

Cheers
Axel


#11

Hi Axel,

Ok great.

Has a new version of the non docker installation been released too?

Best wishes

Paul


#12

Yes, the source installation has also been updated to 2.3.2.