Possible setup instructions for a docker based quantecon about lecture-source-jl HOT 7 CLOSED

I am more worried about the compilation overhead of the DifferentialEquations for the interactive one. If there is any change to their packages, at that point it can make a couple-dozen compilations necessary.

from lecture-source-jl.

The other thing is: do we tag the new QELP to be v0.3.1 instead of v0.4? It may make sense that these tweaks are only used for minor releases...

And I think we should also then push the changed v0.3.1 to the syzygy build, so that users have whatever is the latest after the ] up without a whole bunch of recompilation.

Of course, that means we add it to the syzygy, rather than replacing the current one. i.e. both v0.3 and v0.3.1 are instnatiated int he image.

from lecture-source-jl.

OK @jlperla I think this is complete. See the UBCEcon repo.

Each image comes with our usual set of packages (i.e., Expectations Parameters Plots InstantiateFromURL QuantEcon IJulia DifferentialEquations Interpolations...).

The headless image is about 3 GBs, with roughly 1.7 coming from our installed/precompiled packages, 0.5 coming from the base Julia image, and 0.5 coming from the Linux dependencies necessary for plots. The additional 0.3 is Jupinx/Miniconda.

The interactive image is about 6 GBs, with about 0.5 coming from our packages (smaller most likely because of pre-existing resources), 0.5 coming from Conda dependencies for R, 0.3 from Julia itself, and the rest from the base image. This is bigger than the one Jasmine was experimenting with, because we based it off of the full jupyter/datascience-notebook instead of picking and choosing.

from lecture-source-jl.

Are we using the InstantiateFromURL here to install/precompile all of QELP? Remember that we want this to work with them precompiled out of the box to be compatible with the lecture notes.

But before we do that... lets try to get the image size down. I think we should do the following:

• Make a version of the QELP v0.4 where we remove DifferentialEquations, Dierckx, Interact, DiffEqCallbacks
• Add in LazyArrays, FillArrays as well, but these are dependencies of banded, so it won't add anything
• Do a general ] up afterwards.
• We should check that the manifest.toml is significantly shorter, and we can test with the ECON_628 notebooks to make sure there isn't any issue after we move to v0.4
• If there are a few lecture notes that use Dierckx, people can add it as required.
• Remove the DifferentialEquations DiffEqBase from the headless and interactive ones as well
• Add in the using InstantiateFromURL; activate_github("QuantEcon/QELP", "v0.4") or whatever into both of the images
• We should then have separate headless-diffeq and interactive-diffeq images which use these other two as bases, but simply do a add DifferentialEquations DiffEqBase DiffEqCallbacks Interact Sundials; precompile to them. This gives an option for a precompiled, massive image with all the diffeq stuff in it.
  • I am not sure if you can do this prior to putting the first two as official docker images on the QuantEcon docker build (because we use it as a base image). We can get the login info from Matt, although it is possible jasmine already has it.

from lecture-source-jl.

OK, sounds good. We can also remove R stuff and try and delete unnecessary cache files.

from lecture-source-jl.

All sounds good to me. And agree that we shouldn’t have everything be a minor release.

from lecture-source-jl.

Closing this because we now have the QuantEcon/docker#2

from lecture-source-jl.