Theano compilation error when using marginal Gaussian processes

[fonnesbeck]

When using marginal Gaussian processes, parameterizing the marginal likelihood such as the following:

    y_ = gp.marginal_likelihood("y", X=X, Xu=Xu, y=y, noise=σ)

causes a compilation failure in Theano:

Exception: ("Compilation failed (return status=1): /Users/cfonnesbeck/.theano/compiledir_Darwin-19.0.0-x86_64-i386-64bit-i386-3.7.3-64/tmp5ctmpymk/mod.cpp:510:27: error: non-constant-expression cannot be narrowed from type 'npy_intp' (aka 'long') to 'int' in initializer list [-Wc++11-narrowing].     int init_totals[2] = {V3_n0, V3_n1};.                           ^~~~~. /Users/cfonnesbeck/.theano/compiledir_Darwin-19.0.0-x86_64-i386-64bit-i386-3.7.3-64/tmp5ctmpymk/mod.cpp:510:27: note: insert an explicit cast to silence this issue.     int init_totals[2] = {V3_n0, V3_n1};.                           ^~~~~.                           static_cast<int>( ). /Users/cfonnesbeck/.theano/compiledir_Darwin-19.0.0-x86_64-i386-64bit-i386-3.7.3-64/tmp5ctmpymk/mod.cpp:510:34: error: non-constant-expression cannot be narrowed from type 'npy_intp' (aka 'long') to 'int' in initializer list [-Wc++11-narrowing].     int init_totals[2] = {V3_n0, V3_n1};.                                  ^~~~~. /Users/cfonnesbeck/.theano/compiledir_Darwin-19.0.0-x86_64-i386-64bit-i386-3.7.3-64/tmp5ctmpymk/mod.cpp:510:34: note: insert an explicit cast to silence this issue.     int init_totals[2] = {V3_n0, V3_n1};.                                  ^~~~~.                                  static_cast<int>( ). /Users/cfonnesbeck/.theano/compiledir_Darwin-19.0.0-x86_64-i386-64bit-i386-3.7.3-64/tmp5ctmpymk/mod.cpp:522:9: error: non-constant-expression cannot be narrowed from type 'ssize_t' (aka 'long') to 'int' in initializer list [-Wc++11-narrowing].         V3_stride0, V3_stride1, .         ^~~~~~~~~~. /Users/cfonnesbeck/.theano/compiledir_Darwin-19.0.0-x86_64-i386-64bit-i386-3.7.3-64/tmp5ctmpymk/mod.cpp:522:9: note: insert an explicit cast to silence this issue.         V3_stride0, V3_stride1, .         ^~~~~~~~~~.         static_cast<int>( ). /Users/cfonnesbeck/.theano/compiledir_Darwin-19.0.0-x86_64-i386-64bit-i386-3.7.3-64/tmp5ctmpymk/mod.cpp:522:21: error: non-constant-expression cannot be narrowed from type 'ssize_t' (aka 'long') to 'int' in initializer list [-Wc++11-narrowing].         V3_stride0, V3_stride1, .                     ^~~~~~~~~~. /Users/cfonnesbeck/.theano/compiledir_Darwin-19.0.0-x86_64-i386-64bit-i386-3.7.3-64/tmp5ctmpymk/mod.cpp:522:21: note: insert an explicit cast to silence this issue.         V3_stride0, V3_stride1, .                     ^~~~~~~~~~.                     static_cast<int>( ). /Users/cfonnesbeck/.theano/compiledir_Darwin-19.0.0-x86_64-i386-64bit-i386-3.7.3-64/tmp5ctmpymk/mod.cpp:524:1: error: non-constant-expression cannot be narrowed from type 'ssize_t' (aka 'long') to 'int' in initializer list [-Wc++11-narrowing]. V1_stride0, V1_stride1. ^~~~~~~~~~. /Users/cfonnesbeck/.theano/compiledir_Darwin-19.0.0-x86_64-i386-64bit-i386-3.7.3-64/tmp5ctmpymk/mod.cpp:524:1: note: insert an explicit cast to silence this issue. V1_stride0, V1_stride1. ^~~~~~~~~~. static_cast<int>( ). /Users/cfonnesbeck/.theano/compiledir_Darwin-19.0.0-x86_64-i386-64bit-i386-3.7.3-64/tmp5ctmpymk/mod.cpp:524:13: error: non-constant-expression cannot be narrowed from type 'ssize_t' (aka 'long') to 'int' in initializer list [-Wc++11-narrowing]. V1_stride0, V1_stride1.             ^~~~~~~~~~. /Users/cfonnesbeck/.theano/compiledir_Darwin-19.0.0-x86_64-i386-64bit-i386-3.7.3-64/tmp5ctmpymk/mod.cpp:524:13: note: insert an explicit cast to silence this issue. V1_stride0, V1_stride1.             ^~~~~~~~~~.             static_cast<int>( ). 6 errors generated.. ", '[Elemwise{mul,no_inplace}(TensorConstant{[[ 0.9442 ..78980556]]}, <TensorType(float64, (True, True))>)]')

Looks like a numpy-related issue; mine is 1.17.3.

Versions and main components

• PyMC3 Version: 3.7 from current master
• Theano Version: 1.0.4
• Python Version: 3.7
• Operating system: macOS
• How did you install PyMC3: pip (from master)

[fonnesbeck]

Downgrading numpy to 1.16 does not seem to help.

[fonnesbeck]

A work-around is to have theano ignore the compilation errors:

theano.config.gcc.cxxflags = "-Wno-c++11-narrowing"

[tylercroberts]

If you're on Windows, setting

theano.config.mode = 'FAST_COMPILE' 

fixed it for me. I'm using numpy==1.16.5 and theano==1.0.4, same pymc3 version.

From here, it seems to be an issue with how theano interacts with the multiprocessing library.

[ColCarroll]

Should we set this to a default? @fonnesbeck

[fonnesbeck]

Seems like the expedient way to get fixes out to users quickly.