How to use lab as a python module

[johnholl]

Hi,

I built lab on Ubuntu 14.04. The python module tests all pass. I'm a little confused about importing deepmind_lab and running experiments with it in python. I apologize if this is due to my lack of general python/bazel knowledge. My understanding is that the process of experimentation with deepmind_lab is:

1. create a python file experiment.py. import deepmind_lab and use it in the experiment.
2. add a py_binary entry in the BUILD file for bazel named "experiment"
3. perform bazel run :experiment

Is this correct? And, is there any way to instead run experiments directly with python experiment.py?

[tkoeppe]

Hi - you would need to make your Python program depend on the module, something like this:

py_binary(
    name = "foo",
    srcs = ["foo.py"],
    deps = [":deepmind_lab.so"],
)

Inside your Python code, import the lab module in the same way the demo code does it.

[johnholl]

My understanding is that after doing what you're saying, I would run the program with the command bazel run :foo and not python foo.py. My question is: what do I need to do to run python foo.py, or is lab designed to not be able to run in this way. Initially after downloading and building deepmind_lab.so the command

python random_agent.py

will produce

ImportError: No module named deepmind_lab

I tried rectifying this by adding the bazel-bin directory to PYTHONPATH. This lets me import deepmind_lab, but attempting to create a lab environment ie
lab = deepmind_lab.Lab('demo_map')

produces the error

./libdmlab.so: cannot open shared object file: No such file or directory RuntimeError: Failed to connect RL API.

This runtime error doesn't happen if the python interpreter is called inside the bazel-bin directory. Eventually, after creating symbolic links to the files in bazel-bin and creating a baselab folder containing assets, I managed to create a lab environment without error in a python interpreter. However

lab.reset()

produces a popup error message that says "VM_Create on UI failed. See "~/.deepmind_lab/baselab/crashlog.txt" for details."

I read through the crashlog but couldn't figure what was going on.

[tkoeppe]

Does the reset error happen when you use bazel run?

I think you have to use bazel run in order to get everything set up right. Alternatively, you can run bazel-bin/foo from your project root. But I don't think directly calling python will work.

[johnholl]

No, there's no error when doing bazel run. I ask about running it with python foo.py as I have an existing project (A3C implementation) which I'd like to interface with lab and it'd be much easier to do it that way, but it sounds like this might not be doable.

[tkoeppe]

No, unfortunately Bazel does not really lend itself to exporting reusable library components. This is a known issue, but I'm afraid I don't have a good solution for it right now.

[tkoeppe]

(Try ldd on the DSO; I think it may be linked against other parts of the Bazel workspace which you would also need to copy out.)

[johnholl]

ldd deepmind_lab.so outputs

linux-vdso.so.1 =>  (0x00007fff581cb000)
libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007fd14e116000)
libstdc++.so.6 => /usr/lib/x86_64-linux-gnu/libstdc++.so.6 (0x00007fd14de04000)
libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007fd14dafe000)
libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007fd14d8e7000)
libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007fd14d522000)
/lib64/ld-linux-x86-64.so.2 (0x00007fd14e528000)

[tkoeppe]

Try also the other DSO that's loaded by dmlab_so_loader.cc.

[johnholl]

ldd deepmind_lab.so produces rather more:

linux-vdso.so.1 => (0x00007ffc58bbf000) libOSMesa.so.6 => /usr/lib/x86_64-linux-gnu/libOSMesa.so.6 (0x00007fe840058000) libSDL2-2.0.so.0 => /usr/lib/x86_64-linux-gnu/libSDL2-2.0.so.0 (0x00007fe83fd52000) liblua5.1.so.0 => /usr/lib/x86_64-linux-gnu/liblua5.1.so.0 (0x00007fe83fb25000) libstdc++.so.6 => /usr/lib/x86_64-linux-gnu/libstdc++.so.6 (0x00007fe83f813000) libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007fe83f50c000) libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007fe83f2f5000) libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007fe83ef30000) libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007fe83ed11000) libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007fe83eb0d000) libasound.so.2 => /usr/lib/x86_64-linux-gnu/libasound.so.2 (0x00007fe83e81d000) libpulse-simple.so.0 => /usr/lib/x86_64-linux-gnu/libpulse-simple.so.0 (0x00007fe83e618000) libpulse.so.0 => /usr/lib/x86_64-linux-gnu/libpulse.so.0 (0x00007fe83e3cf000) libX11.so.6 => /usr/lib/x86_64-linux-gnu/libX11.so.6 (0x00007fe83e09a000) libXext.so.6 => /usr/lib/x86_64-linux-gnu/libXext.so.6 (0x00007fe83de87000) libXcursor.so.1 => /usr/lib/x86_64-linux-gnu/libXcursor.so.1 (0x00007fe83dc7d000) libXinerama.so.1 => /usr/lib/x86_64-linux-gnu/libXinerama.so.1 (0x00007fe83da7a000) libXi.so.6 => /usr/lib/x86_64-linux-gnu/libXi.so.6 (0x00007fe83d869000) libXrandr.so.2 => /usr/lib/x86_64-linux-gnu/libXrandr.so.2 (0x00007fe83d65f000) libXss.so.1 => /usr/lib/x86_64-linux-gnu/libXss.so.1 (0x00007fe83d45b000) libXxf86vm.so.1 => /usr/lib/x86_64-linux-gnu/libXxf86vm.so.1 (0x00007fe83d254000) libwayland-egl.so.1 => /usr/lib/x86_64-linux-gnu/libwayland-egl.so.1 (0x00007fe83d052000) libwayland-client.so.0 => /usr/lib/x86_64-linux-gnu/libwayland-client.so.0 (0x00007fe83ce45000) libwayland-cursor.so.0 => /usr/lib/x86_64-linux-gnu/libwayland-cursor.so.0 (0x00007fe83cc3c000) libxkbcommon.so.0 => /usr/lib/x86_64-linux-gnu/libxkbcommon.so.0 (0x00007fe83ca02000) librt.so.1 => /lib/x86_64-linux-gnu/librt.so.1 (0x00007fe83c7fa000) /lib64/ld-linux-x86-64.so.2 (0x0000564811452000) libpulsecommon-4.0.so => /usr/lib/x86_64-linux-gnu/pulseaudio/libpulsecommon-4.0.so (0x00007fe83c592000) libjson-c.so.2 => /lib/x86_64-linux-gnu/libjson-c.so.2 (0x00007fe83c387000) libdbus-1.so.3 => /lib/x86_64-linux-gnu/libdbus-1.so.3 (0x00007fe83c141000) libxcb.so.1 => /usr/lib/x86_64-linux-gnu/libxcb.so.1 (0x00007fe83bf22000) libXrender.so.1 => /usr/lib/x86_64-linux-gnu/libXrender.so.1 (0x00007fe83bd18000) libXfixes.so.3 => /usr/lib/x86_64-linux-gnu/libXfixes.so.3 (0x00007fe83bb11000) libffi.so.6 => /usr/lib/x86_64-linux-gnu/libffi.so.6 (0x00007fe83b909000) libwrap.so.0 => /lib/x86_64-linux-gnu/libwrap.so.0 (0x00007fe83b6fe000) libsndfile.so.1 => /usr/lib/x86_64-linux-gnu/libsndfile.so.1 (0x00007fe83b496000) libasyncns.so.0 => /usr/lib/x86_64-linux-gnu/libasyncns.so.0 (0x00007fe83b290000) libXau.so.6 => /usr/lib/x86_64-linux-gnu/libXau.so.6 (0x00007fe83b08b000) libXdmcp.so.6 => /usr/lib/x86_64-linux-gnu/libXdmcp.so.6 (0x00007fe83ae85000) libnsl.so.1 => /lib/x86_64-linux-gnu/libnsl.so.1 (0x00007fe83ac6b000) libFLAC.so.8 => /usr/lib/x86_64-linux-gnu/libFLAC.so.8 (0x00007fe83aa39000) libvorbisenc.so.2 => /usr/lib/x86_64-linux-gnu/libvorbisenc.so.2 (0x00007fe83a56a000) libvorbis.so.0 => /usr/lib/x86_64-linux-gnu/libvorbis.so.0 (0x00007fe83a33d000) libogg.so.0 => /usr/lib/x86_64-linux-gnu/libogg.so.0 (0x00007fe83a133000) libresolv.so.2 => /lib/x86_64-linux-gnu/libresolv.so.2 (0x00007fe839f18000)

Also, I was thinking it would be easier to give this up and directly run the binaries that bazel creates, however I run into the exact same issues listed above when I try to run the binaries with ./bazel-bin/foo! So the only way I am able to run anything is using bazel run :foo, but bazel only allows one process to be run at a time which makes running things this way not ideal.

[miyosuda]

@johnholl
I'm not sure whether this helps or not, but I tried UNREAL algorithm based on A3C with lab.

https://github.com/miyosuda/unreal

I put project dir inside lab dir and put another BUILD file for it.
First I tried to call lab environment from outside, but it was complicated so I decided to do it like this.

[johnholl]

Thanks @miyosuda, that's definitely very helpful. Also, another person in my lab, @junhyukoh, figured out that /bazel-bin/random_agent.runfiles/org_deepmind_lab is where all the necessary files reside. Running a python interpreter from here allows one to import deepmind lab, create a lab environment, and reset it without any problems.

[tkoeppe]

Ah yes, that's right: there's a wrapper script in bazel-bin/foo, but the complete directory structure as created by the various dependencies is in bazel-bin/foo.runfiles/..., so relative to that location all the files are where they should be.

[arpit15]

@tkoeppe the solution proposed by @miyosuda works perfectly fine. However I think the better way would be to let bazel handle it automatically. I am trying to get a working example which uses the bazel tools to get lab as external dependency. However I am facing some problems.

I get following error

ERROR: ~/.cache/bazel/_bazel_noone/ae6e9d00e8c783d8279db324d79ccc60/external/org_deepmind_lab/BUILD:432:1: no such package '@zlib_archive//': error loading package 'external': The repository named 'zlib_archive' could not be resolved and referenced by '@org_deepmind_lab//:bspc'.
ERROR: Analysis of target '//:random_agent' failed; build aborted.
INFO: Elapsed time: 13.452s

`
$tree

my-project
├── BUILD
├── python.BUILD
├── random_agent.py
└── WORKSPACE
0 directories, 4 files
$cat WORKSPACE

workspace(name = "maze_a3c")

new_local_repository(
name = "python_system",
build_file = "python.BUILD",
path = "/usr",
)

git_repository(
name = "org_deepmind_lab",
remote = "https://github.com/deepmind/lab.git",
commit = "9af8f87376c3418dd40ade46465d677c0b96f39e"

$cat BUILD

py_binary(
name = "random_agent",
srcs = ["random_agent.py"],
data = ["@org_deepmind_lab//:deepmind_lab.so"],
main = "random_agent.py",

`

Kindly suggest what is the correct way to add bazel repository as external dependency and also correctly build it. As this pertains more to bazel build system I have asked the question there as well. Another similar issue in bazel.

[tkoeppe]

@arpit15: I'm afraid I don't have an answer for you.

I've discussed this with the Bazel team in the past, and I think we all agree that Bazel is not very well suited at the moment to make exportable builds of reusable libraries. Bazel works well for a single-root project, where it gives you a very convenient mechanism to assemble dependencies of both compile-time code and runtime assets -- but it only works if you stay within the confines of the system.

My understanding is that the Bazel team is working on making Bazel more flexible, and I'd encourage you to keep up with the developments there. But as far as the immediate present is concerned, I am thinking that it might be useful to set up an alternative build system if you want something exportable -- but I don't know what that would look like.

[RylanSchaeffer]

@tkoeppe , do you any updates on this issue?

[tkoeppe]

@RylanSchaeffer: I think we have some code in the works to produce a pip package. I hope we'll get to push that soon.

[RylanSchaeffer]

@tkoeppe Awesome! Do you have a rough estimate of when "soon" will be?

[tkoeppe]

@RylanSchaeffer: No hard guarantees, but hopefully soon.

[tkoeppe]

We didn't put this into the recent release, but it should make it into HEAD very soon.

[tkoeppe]

This should be available as of 0a0638c.

[RylanSchaeffer]

@tkoeppe , I finally returned to this issue and I ran into a problem following the instructions. Specifically,

sudo bazel build python/pip_package:build_pip_package
Extracting Bazel installation...
............
INFO: Analysed target //python/pip_package:build_pip_package (48 packages loaded).
INFO: Found 1 target...
ERROR: /localhome/lab/deepmind/lua/BUILD:140:1: C++ compilation of rule '//deepmind/lua:push_script' failed (Exit 1)
In file included from ./deepmind/lua/push_script.h:26:0,
                 from deepmind/lua/push_script.cc:19:
./deepmind/lua/lua.h:25:21: fatal error: lauxlib.h: No such file or directory
compilation terminated.
Target //python/pip_package:build_pip_package failed to build
Use --verbose_failures to see the command lines of failed build steps.
INFO: Elapsed time: 20.730s, Critical Path: 2.02s
FAILED: Build did NOT complete successfully

Do you know what might have caused this?

[tkoeppe]

@RylanSchaeffer: On which platform is this? You may have to edit lua.BUILD as described somewhere in https://github.com/deepmind/lab/blob/master/docs/users/build.md to match your system's library paths.