macos - 在 MacOS 上使用 Julia 0.7.0 的 TensorFlow.jl 出现分段错误

Question

我正在尝试在 MacOS 上为 Julia v0.7.0 安装 TensorFlow.jl。我安装了以下软件包：

(v0.7) pkg> status
Status `~/.julia/environments/v0.7/Project.toml`
[336ed68f] CSV v0.3.1
[a93c6f00] DataFrames v0.13.1
[7073ff75] IJulia v1.12.0
[438e738f] PyCall v1.18.4
[1d978283] TensorFlow v0.10.2

运行以下 MWE 时

using TensorFlow
sess = Session()
feature_columns = placeholder(Float32)
target_columns = placeholder(Float32)
m=Variable(0.05)
b=Variable(0.0)
y=m.*feature_columns+b
loss=reduce_sum((target_columns - y).^2)
learning_rate=0.0000001;
my_optimizer=train.minimize(train.GradientDescentOptimizer(learning_rate), loss)

第一次，我收到错误消息

On worker 2:
LoadError: ArgumentError: Package MacroTools not found in current path:
- Run `Pkg.add("MacroTools")` to install the MacroTools package.

通过包管理器添加MacroTools.jl并重新运行示例后，Julia 内核因分段错误而崩溃：

signal (11): Segmentation fault: 11
in expression starting at no file:0
TF_GraphGetTensorNumDims at /Users/sd/.julia/packages/TensorFlow/A6TdG/deps/usr/bin/libtensorflow.dylib (unknown line)
get_tensor_num_dims at /Users/sd/.julia/packages/TensorFlow/A6TdG/src/core.jl:25
get_shape at /Users/sd/.julia/packages/TensorFlow/A6TdG/src/core.jl:1823
show at /Users/sd/.julia/packages/TensorFlow/A6TdG/src/show.jl:50
show at ./sysimg.jl:195
jl_fptr_trampoline at /Users/osx/buildbot/slave/package_osx64/build/src/gf.c:1829
display at /Users/osx/buildbot/slave/package_osx64/build/usr/share/julia/stdlib/v0.7/REPL/src/REPL.jl:131
display at /Users/osx/buildbot/slave/package_osx64/build/usr/share/julia/stdlib/v0.7/REPL/src/REPL.jl:135
display at ./multimedia.jl:287
#7 at /Users/sd/.julia/packages/Media/Lrdeg/src/compat.jl:28
hookless at /Users/sd/.julia/packages/Media/Lrdeg/src/compat.jl:14
render at /Users/sd/.julia/packages/Media/Lrdeg/src/compat.jl:27
render at /Users/sd/.julia/packages/Media/Lrdeg/src/system.jl:160
display at /Users/sd/.julia/packages/Media/Lrdeg/src/compat.jl:9
display at ./multimedia.jl:287
jl_fptr_trampoline at /Users/osx/buildbot/slave/package_osx64/build/src/gf.c:1829
jl_apply at /Users/osx/buildbot/slave/package_osx64/build/src/./julia.h:1538 [inlined]
jl_f__apply at /Users/osx/buildbot/slave/package_osx64/build/src/builtins.c:563
jl_f__apply_latest at /Users/osx/buildbot/slave/package_osx64/build/src/builtins.c:601
#invokelatest#1 at ./essentials.jl:691 [inlined]
invokelatest at ./essentials.jl:690 [inlined]
print_response at /Users/osx/buildbot/slave/package_osx64/build/usr/share/julia/stdlib/v0.7/REPL/src/REPL.jl:154
unknown function (ip: 0x109ead87e)
print_response at /Users/osx/buildbot/slave/package_osx64/build/usr/share/julia/stdlib/v0.7/REPL/src/REPL.jl:139
unknown function (ip: 0x109ead2ab)
do_respond at /Users/osx/buildbot/slave/package_osx64/build/usr/share/julia/stdlib/v0.7/REPL/src/REPL.jl:708
unknown function (ip: 0x109e96124)
jl_apply at /Users/osx/buildbot/slave/package_osx64/build/src/./julia.h:1538 [inlined]
jl_f__apply at /Users/osx/buildbot/slave/package_osx64/build/src/builtins.c:563
jl_f__apply_latest at /Users/osx/buildbot/slave/package_osx64/build/src/builtins.c:601
#invokelatest#1 at ./essentials.jl:691 [inlined]
invokelatest at ./essentials.jl:690 [inlined]
run_interface at /Users/osx/buildbot/slave/package_osx64/build/usr/share/julia/stdlib/v0.7/REPL/src/LineEdit.jl:2261
run_frontend at /Users/osx/buildbot/slave/package_osx64/build/usr/share/julia/stdlib/v0.7/REPL/src/REPL.jl:1029
run_repl at /Users/osx/buildbot/slave/package_osx64/build/usr/share/julia/stdlib/v0.7/REPL/src/REPL.jl:191
#831 at ./logging.jl:311
jfptr_#831_5851.clone_1 at /Applications/Programmieren/Julia-0.7.app/Contents/Resources/julia/lib/julia/sys.dylib (unknown line)
jl_apply at /Users/osx/buildbot/slave/package_osx64/build/src/./julia.h:1538 [inlined]
jl_f__apply at /Users/osx/buildbot/slave/package_osx64/build/src/builtins.c:563
jl_f__apply_latest at /Users/osx/buildbot/slave/package_osx64/build/src/builtins.c:601
#invokelatest#1 at ./essentials.jl:691 [inlined]
invokelatest at ./essentials.jl:690 [inlined]
macro expansion at ./logging.jl:308 [inlined]
run_main_repl at ./client.jl:340
exec_options at ./client.jl:252
_start at ./client.jl:432
true_main at /Applications/Programmieren/Julia-0.7.app/Contents/Resources/julia/bin/julia (unknown line)
main at /Applications/Programmieren/Julia-0.7.app/Contents/Resources/julia/bin/julia (unknown line)
Allocations: 66064400 (Pool: 66051866; Big: 12534); GC: 149

在其他一些情况下，我收到了错误消息

On worker 2: The Python TensorFlow package could not be imported. You must install Python TensorFlow before using this package. error at ./error.jl:33

或者

UndefVarError: py_gradients not defined

不幸的是，在重新安装所有内容并从头开始后，我无法重现它们。我认为这些错误与PyCall.jlConda 的当前问题有关。

关于如何让包工作有什么想法吗？

标签： macostensorflowjulia