LinalgRefactor - specialpurpose #3767

OXPHOS · 2017-04-10T05:14:23Z

No description provided.

OXPHOS · 2017-04-10T16:38:34Z

src/shogun/mathematics/linalg/LinalgBackendBase.h

+	#undef BACKEND_GENERIC_SOFTMAX
+
+ /**
+	* Wrapper method of squared error method.


OXPHOS · 2017-04-10T16:38:59Z

src/shogun/mathematics/linalg/LinalgBackendEigen.h

@@ -127,6 +133,16 @@ class LinalgBackendEigen : public LinalgBackendBase
 	DEFINE_FOR_NON_INTEGER_PTYPE(BACKEND_GENERIC_CHOLESKY_SOLVER, SGMatrix)
 	#undef BACKEND_GENERIC_CHOLESKY_SOLVER

+ /** Implementation of @see linalg::cross_entropy */
+ #define BACKEND_GENERIC_CROSS_ENTROPY(Type, Container) \


OXPHOS · 2017-04-10T16:56:18Z

src/shogun/mathematics/linalg/LinalgBackendEigen.h

+		multiply_by_rectified_linear_derivative_impl(a, result); \
+	}
+	DEFINE_FOR_REAL_PTYPE(BACKEND_GENERIC_MULTIPLY_BY_RECTIFIED_LINEAR_DERIV, SGMatrix)
+	#undef BACKEND_GENERIC_MULTIPLY_BY_RECTIFIED_LINEAR_DERIV


non integer real.
Interger doesn't really make sense here.

OXPHOS · 2017-04-10T16:56:56Z

src/shogun/mathematics/linalg/LinalgBackendEigen.h

+		softmax_impl(a); \
+	}
+	DEFINE_FOR_REAL_PTYPE(BACKEND_GENERIC_SOFTMAX, SGMatrix)
+	#undef BACKEND_GENERIC_SOFTMAX


Non integer real

OXPHOS · 2017-04-10T16:57:30Z

src/shogun/mathematics/linalg/LinalgBackendEigen.h

+		typename SGMatrix<T>::EigenMatrixXtMap a_eig = a;
+
+		auto max = a_eig.maxCoeff();
+		for (index_t j = 0; j < a.num_cols; ++j) //a.num_cols; ++j)


OXPHOS · 2017-04-10T16:57:53Z

src/shogun/mathematics/linalg/LinalgBackendViennaCL.h

@@ -85,6 +85,17 @@ class LinalgBackendViennaCL : public LinalgBackendGPUBase
 	DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_ADD, SGMatrix)
 	#undef BACKEND_GENERIC_ADD

+
+	/** Implementation of @see linalg::cross_entropy */
+  #define BACKEND_GENERIC_CROSS_ENTROPY(Type, Container) \


OXPHOS · 2017-04-10T16:59:23Z

src/shogun/mathematics/linalg/LinalgBackendViennaCL.h

+
+		result.gpu_ptr = std::shared_ptr<GPUMemoryBase<T>>(
+			result_gpu->clone_vector(result_gpu,a.num_rows*a.num_cols));
+}


OXPHOS · 2017-04-10T17:01:09Z

src/shogun/mathematics/linalg/internal/implementation/SpecialPurpose.h

@@ -444,6 +444,8 @@ struct cross_entropy<Backend::EIGEN3,Matrix>
 	}
 };

+
+
 #ifdef HAVE_VIENNACL


OXPHOS · 2017-05-05T22:42:06Z

Hi @geektoni Could you help me look at my format-check results?
I brew installed clang-format and did git clang-format --commit 0b713865cf064e6a4fdc80cf4aae6c511455b34e --binary /usr/local/bin/clang-format as suggested by travis. I feel like some indentations are moved randomly and so are some back-slashes . Am I missing something?

geektoni · 2017-05-06T09:16:44Z

Hi @OXPHOS. What I found is that:

Travis terminal cannot show very well tabs indentation (it's the reason behind the fact that all the backslashes seem to have been moved randomly). If you look at your editor, all the backslashes should be perfectly aligned on the right-most side (mine does);
Clang-format evaluate #define without respecting indentation levels (so it will delete all the tabs before it). Moreover, other indentations seem weird because clang-format align the macro calls to the public keyword, which means:

namespace b {
	class a {
		public:

#define BEAUTIFUL_MACRO() /* This won't respect namespace and class indent*/
	void method(...) \
	{
	....	\
	}
		OTHER_MACRO_CALL() /* this is aligned to public keyword below */
#undef BEAUTIFUL_MACRO	/* The same as #define */
	};
}

Unfortunately, I don't think this last behaviour can be fixed easily.

I also see you applied the changes suggested by clang-format, but Travis still fails. Could you please run the same command again? Could you tell me also the clang-format version that you're using? Thanks :)

OXPHOS · 2017-06-05T04:33:53Z

@geektoni Sorry about my late reply..I was away for the last month.

We were talking about that clang-format cannot handle macros very well. My current solution is guard macros with \\ clang-format off - if the backslash alignment is not a hard requirement.

Also, I think that the diffs on travis after I performed clang-format on my machine. My clang-format version is : clang-format version 5.0.0 (tags/google/stable/2017-03-17). I did a brief search and only found information about installing clang-format-3.8 on ubuntu while I am using macbook.

geektoni · 2017-06-05T07:50:14Z

Sorry about my late reply..I was away for the last month.

@OXPHOS no problem!

We were talking about that clang-format cannot handle macros very well. My current solution is guard macros with \ clang-format off - if the backslash alignment is not a hard requirement.

Backslash alignment is just to make the code prettier, so it is fine not doing it. However, we should use \\ clang-format off carefully, since if we begin to "protect" code too often, clang-format will become useless ;)

Also, I think that the diffs on travis after I performed clang-format on my machine. My clang-format version is : clang-format version 5.0.0 (tags/google/stable/2017-03-17). I did a brief search and only found information about installing clang-format-3.8 on ubuntu while I am using macbook.

What do you mean here? It seems that a piece of information is missing. Are you talking about installing clang-format-3.8 on OSX?

OXPHOS · 2017-06-06T05:07:26Z

What do you mean here? It seems that a piece of information is missing. Are you talking about installing clang-format-3.8 on OSX?

@geektoni Yes I was wondering how to install clang-format-3.8 on OSX.
But I might find another way by doing the format checking in Docker. Let me see how it works on Travis.

karlnapf · 2017-06-16T15:40:06Z

Can we push this in?

OXPHOS added this to In Progress in Linalg refactor Apr 10, 2017

OXPHOS commented Apr 10, 2017

View reviewed changes

OXPHOS force-pushed the linalg_specialpurpose branch from 89211a2 to 33517d0 Compare May 4, 2017 19:47

OXPHOS force-pushed the linalg_specialpurpose branch 2 times, most recently from a5686eb to 60f81d8 Compare June 5, 2017 04:27

OXPHOS force-pushed the linalg_specialpurpose branch from 60f81d8 to 1bcd948 Compare June 6, 2017 04:58

OXPHOS force-pushed the linalg_specialpurpose branch 2 times, most recently from 04b35a1 to b3289d1 Compare June 6, 2017 17:38

Linalg special purpose - CPU

df1ecf3

OXPHOS force-pushed the linalg_specialpurpose branch from b3289d1 to df1ecf3 Compare June 27, 2017 13:27

OXPHOS added 3 commits June 27, 2017 16:09

Linalg special purpose - GPU

652ee0d

Format check

85cb5a7

rm old special purpose

d552c0a

OXPHOS merged commit 6b0fd29 into shogun-toolbox:develop Jun 27, 2017

OXPHOS moved this from In Progress to DONE in Linalg refactor Jun 28, 2017

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

LinalgRefactor - specialpurpose #3767

LinalgRefactor - specialpurpose #3767

OXPHOS commented Apr 10, 2017 •

edited

OXPHOS Apr 10, 2017

OXPHOS Apr 10, 2017

OXPHOS Apr 10, 2017

OXPHOS Apr 10, 2017

OXPHOS Apr 10, 2017

OXPHOS Apr 10, 2017

OXPHOS Apr 10, 2017

OXPHOS Apr 10, 2017

OXPHOS commented May 5, 2017

geektoni commented May 6, 2017

OXPHOS commented Jun 5, 2017

geektoni commented Jun 5, 2017

OXPHOS commented Jun 6, 2017

karlnapf commented Jun 16, 2017

LinalgRefactor - specialpurpose #3767

LinalgRefactor - specialpurpose #3767

Conversation

OXPHOS commented Apr 10, 2017 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

OXPHOS commented May 5, 2017

geektoni commented May 6, 2017

OXPHOS commented Jun 5, 2017

geektoni commented Jun 5, 2017

OXPHOS commented Jun 6, 2017

karlnapf commented Jun 16, 2017

OXPHOS commented Apr 10, 2017 •

edited