card.io-dmz
card.io-dmz copied to clipboard
card-io
models/generated: regenerate number models
@dgoldman-ebay I renamed some of our model files, and I wanted to re-export them so that the file names would match here. However, the export code was broken for our old files (no dropout_rate, no activation) so I had to fix it up. I was thus disconcerted when this diff came out as big as it is, given that this is the same underlying model as before.
Would you mind looking through this diff and confirming that all these changes are expected and desired? Thanks!
The strip model regeneration looks more like the kind of diff I expected.
By the way, have we checked on the performance impact of removing EIGEN_NO_DEBUG?
I'll try to take a look late tonight, or else as soon as possible thereafter.
(Our code generators can now handle N-level models; formerly limited to two(?) levels. The changes related to that are probably responsible for most of the diff.)
Any time this week is fine. I have plenty of other things to fix in the meantime. :)
Hmmm. My money is on this not being right. :) It currently doesn't compile, due to the switch from #include "cv/conv.h" to just #include "conv.h". Switching that back causes an explosion of compilation errors from Eigen. Here's a sample error:
In file included from /Users/jbleechersnyder/src/clearcut/dmz/card.io-dmz/dmz_all.cpp:15:
In file included from /Users/jbleechersnyder/src/clearcut/dmz/card.io-dmz/./cv/sobel.cpp:9:
In file included from /Users/jbleechersnyder/src/clearcut/dmz/card.io-dmz/eigen.h:25:
In file included from /Users/jbleechersnyder/src/clearcut/dmz/card.io-dmz/Eigen/Core:284:
/Users/jbleechersnyder/src/clearcut/dmz/card.io-dmz/Eigen/src/Core/Assign.h:493:32: error: no member named
'YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY' in
'Eigen::internal::static_assertion<false>'
EIGEN_STATIC_ASSERT(SameType,YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYP...
^
/Users/jbleechersnyder/src/clearcut/dmz/card.io-dmz/Eigen/src/Core/util/StaticAssert.h:114:65: note: expanded from macro
'EIGEN_STATIC_ASSERT'
if (Eigen::internal::static_assertion<bool(CONDITION)>::MSG) {}
^
/Users/jbleechersnyder/src/clearcut/dmz/card.io-dmz/Eigen/src/Core/PlainObjectBase.h:414:20: note: in instantiation of function template
specialization 'Eigen::DenseBase<Eigen::Matrix<float, 8, 40, 1, 8, 40>
>::lazyAssign<Eigen::CwiseUnaryOp<std::__1::pointer_to_unary_function<double, double>, const Eigen::Matrix<float, 8, 40, 1, 8, 40> >
>' requested here
return Base::lazyAssign(other.derived());
^
/Users/jbleechersnyder/src/clearcut/dmz/card.io-dmz/Eigen/src/Core/Assign.h:520:97: note: in instantiation of function template
specialization 'Eigen::PlainObjectBase<Eigen::Matrix<float, 8, 40, 1, 8, 40>
>::lazyAssign<Eigen::CwiseUnaryOp<std::__1::pointer_to_unary_function<double, double>, const Eigen::Matrix<float, 8, 40, 1, 8, 40> >
>' requested here
static EIGEN_STRONG_INLINE Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.derived()); }
^
/Users/jbleechersnyder/src/clearcut/dmz/card.io-dmz/Eigen/src/Core/PlainObjectBase.h:621:69: note: in instantiation of member function
'Eigen::internal::assign_selector<Eigen::Matrix<float, 8, 40, 1, 8, 40>,
Eigen::CwiseUnaryOp<std::__1::pointer_to_unary_function<double, double>, const Eigen::Matrix<float, 8, 40, 1, 8, 40> >, false,
false>::run' requested here
return internal::assign_selector<Derived,OtherDerived,false>::run(this->derived(), other.derived());
^
/Users/jbleechersnyder/src/clearcut/dmz/card.io-dmz/Eigen/src/Core/PlainObjectBase.h:606:102: note: in instantiation of function template
specialization 'Eigen::PlainObjectBase<Eigen::Matrix<float, 8, 40, 1, 8, 40>
>::_set_noalias<Eigen::CwiseUnaryOp<std::__1::pointer_to_unary_function<double, double>, const Eigen::Matrix<float, 8, 40, 1, 8, 40>
> >' requested here
EIGEN_STRONG_INLINE void _set_selector(const OtherDerived& other, const internal::false_type&) { _set_noalias(other); }
^
/Users/jbleechersnyder/src/clearcut/dmz/card.io-dmz/Eigen/src/Core/PlainObjectBase.h:598:7: note: in instantiation of function template
specialization 'Eigen::PlainObjectBase<Eigen::Matrix<float, 8, 40, 1, 8, 40>
>::_set_selector<Eigen::CwiseUnaryOp<std::__1::pointer_to_unary_function<double, double>, const Eigen::Matrix<float, 8, 40, 1, 8, 40>
> >' requested here
_set_selector(other.derived(), typename internal::conditional<static_cast<bool>(int(OtherDerived::Flags) & EvalBeforeAssignin...
^
/Users/jbleechersnyder/src/clearcut/dmz/card.io-dmz/Eigen/src/Core/Matrix.h:172:20: note: in instantiation of function template
specialization 'Eigen::PlainObjectBase<Eigen::Matrix<float, 8, 40, 1, 8, 40>
>::_set<Eigen::CwiseUnaryOp<std::__1::pointer_to_unary_function<double, double>, const Eigen::Matrix<float, 8, 40, 1, 8, 40> > >'
requested here
return Base::_set(other);
^
/Users/jbleechersnyder/src/clearcut/dmz/card.io-dmz/./models/generated/modelc_01266c1b.cpp:1939:23: note: in instantiation of function
template specialization 'Eigen::Matrix<float, 8, 40, 1, 8,
40>::operator=<Eigen::CwiseUnaryOp<std::__1::pointer_to_unary_function<double, double>, const Eigen::Matrix<float, 8, 40, 1, 8, 40> >
>' requested here
accumulated_results = accumulated_results.unaryExpr(std::ptr_fun(tanh));
^
(How does importing just "conv.h" work for the expiry models?)
You're building the DMZ in the context of Cython? I don't think that the expiry models are included in that context; they're included only when building CardIO (SCAN_EXPIRY isn't defined in the Cython case).
For building CardIO, $(SRCROOT)/dmz is one of the header search paths. Possibly that's the magic?
Not sure about the Eigen errors...
FWIW, I just pasted the contents of your new modelc_01266c1b.cpp into my old file of the same name, and tried to build CardIO.
No error re #include "conv.h".
A couple of Eigen errors, though. Sigh.