Add Finite-Diffs (FD Branch) to RALFit #121

as-amd · 2024-08-14T12:58:27Z

This branch enables

* Approximate Jacobian matrix using FD
* Check user-supplied Jacobian using FD


* Update doc
* Update UT
* Add examples

* Adds support for FD in eval_J * Tweaks examples * Add copyright notice banners Change-Id: I2ea721708a323c840f08af2d254b9c68e5f259a7

Change-Id: Ia05dbe4d07d00f66b00fd7555497a3b3d415f34f

Add module to parametrize data types (int, real) in preparation to build LP64/ILP64 and float/double variants It also tweaks some of the examples Change-Id: I8440c86be379839677bf65f5163e9d38cd17bb1f

Also fixes nlls_example2.c bug Change-Id: Iec2ec4983eae41e229472d2bf902258a0a229d1d

Change-Id: I45d8f565b07c4c7330317061db53ed09941f5e3e

* jacobian_setup, * jacobian_calc and * jacobian_free * example to show how to call this framework instead of using the solver's internal mecanism Change-Id: I4eb24ce1cf67a78dceac541066f509657ef3d202

Example to illustrate how to use RALFit FD machinery to estimate a Jacobian matrix outside of the solver Change-Id: I34b81558e6375c8973f6988a65f7708354f3b391

Update and tidy examples Change-Id: I094948f5705f836c7d2aa4905b6081ed6f0c6c06

Added new extended examples Fixed bug in weight size passed in linesearch Change-Id: Ia58f6c8bf74e7da3999b2c69e48fed874c95d112

* tweak examples * UT fd checker * UT fd solver Change-Id: I8f8d4fb8674f7058a453400b1d66449f0ea08c4d

Change-Id: I223cb5906f8853b7b346bef5a7ef18f2304a772e

Change-Id: Icdbf3b11671b17cc9bb2248a3e60ba07d2042630

Fixes ifort compilation error Change-Id: Ib6ece887276e4a2700c79d26d224182edf0d3ec4

Still need to move common chunks to `../common` Change-Id: Id611ab225488814649606de88bc3085c34f20689

Change-Id: Ibf48fa781dda37c4adc9cd581da0512dbc5177c3

* update the docs for python 3 and rtd * Fix warnings, including c issues * fix typos and c signature * Add a docs readme * Fix python warnings/errors

* Typos * Removal of trailing blanks Change-Id: I85304c042b21728edfc544b12fa93b9a0d76eac7

Change-Id: I2f3d96e9af7d878b40a54bc05197fc94a2fae070

tyronerees

Many thanks for this! I've dotted a few comments through the code.

I get a test failure when I run the code locally:

I've got a few more things to check, but I thought it's worth feeding this back from the first pass.

Also note that @AndrewLister-STFC fixed the python installation for us -- that's now merged into master (and doesn't conflict with anything here)

tyronerees · 2024-08-30T12:42:48Z

libRALFit/CMakeLists.txt

@@ -1,10 +1,18 @@
 # Copyright (c) 2016, The Science and Technology Facilities Council (STFC)
 # All rights reserved.
+# Copyright (C) 2024 Advanced Micro Devices, Inc. All rights reserved.
 # CMake project file for libral_nlls
 cmake_minimum_required (VERSION 3.1)


I think this needs to be bumped -- Fortran_PREPROCESS was added in 3.18, so that should (at least) be the minimum.

tyronerees · 2024-08-30T12:43:59Z

libRALFit/CMakeLists.txt

+# Request to run pre-processor as well
+if(NOT WIN32)
+  set(CMAKE_Fortran_FLAGS -cpp)
+else()
+  set(CMAKE_Fortran_PREPROCESS ON)
+endif()


What does CMAKE_Fortran_PREPROCESS do? Is it enough to set this for all architectures -- would that set the right flag for us?

This CMake flag calls CPP on .F90 files and consumes any C-preprocessor directive in Fortran 90 files, so #include (et al) work. We can certainly enable it for all architectures.

tyronerees · 2024-08-30T12:54:27Z

libRALFit/doc/C/howtouse.rst

@@ -202,6 +202,9 @@ It must have the following signature:

   :param status: |eval_J_status|

+If user does not provide the call-back it may be substituted by ``NULL`` and 


I'm not sure this needs to be here as well. I think I'd change must -> may in the description before this, and possibly mention changing to NULL there (although I'm not sure this is needed). This is all said in the calling sequence, and I think that's enough.

Changed in C and Fortran howtouse.rst

tyronerees · 2024-08-30T12:55:00Z

libRALFit/doc/C/howtouse.rst

@@ -229,6 +232,8 @@ It must have the following signature:

   :param status: |eval_Hf_status|

+If user does not provide the call-back it may be substituted by ``NULL``.


Same comment here: changing must -> may is probably enough.

libRALFit/doc/C/howtouse.rst

libRALFit/test/nlls_test.f90

tyronerees · 2024-08-30T14:34:18Z

libRALFit/src/ral_nlls_types.F90

+#ifdef SINGLE_PRECISION
+   Integer, Parameter, Public :: wp = lp
+   Integer, Parameter, Public :: ral_c_real = c_float
+#else
+   Integer, Parameter, Public :: wp = np
+   Integer, Parameter, Public :: ral_c_real = c_double
+#endif
+
+#ifdef BUILD_ILP64
+   ! To build ilp64 also the default integer kind must be set to 8 bytes
+   Integer, Parameter, Public :: ral_c_int = c_int64_t
+#else
+   Integer, Parameter, Public :: ral_c_int = c_int
+#endif
+


I might be missing something, but I'm not sure this is enough. There are lots of calls to lapack and blas in the codes, and the calls to d___ will have to change to s___

It might also be worth documenting how to switch precisions in the installation section of the docs.

I'm glad you brought this up, these changes don't introduce support for single precision it only tries to gather into a single place wp and it's C pair. The support for single precision PR is much, much more extensive! My next PR addresses this topic and will be properly documented.

Suggestion accepted Co-authored-by: Tyrone Rees <[email protected]>

Co-authored-by: Tyrone Rees <[email protected]>

Change-Id: Ic63a9aeacf759fc1e67c05955233b9b92f0a0b3f

Remove duplicated text and include from common file Change-Id: I984b3a7b33239bf282f5395be14d9691f186c790

On some numerical lib/hardware combinations some FD tests where failing. Tolerances are tweaked to account for this variability, also some test with starting iterate are updated to be "closer" to the required solution. Change-Id: I33d97c1591c4c3a509f2873a662a6f546238fc01

Andrew Sajo and others added 19 commits July 10, 2024 17:24

Finite Differences Jacobian

c61ed54

* Adds support for FD in eval_J * Tweaks examples * Add copyright notice banners Change-Id: I2ea721708a323c840f08af2d254b9c68e5f259a7

Tidy output of Fortran example

ca1d041

Change-Id: Ia05dbe4d07d00f66b00fd7555497a3b3d415f34f

Add type module

30e94f5

Add module to parametrize data types (int, real) in preparation to build LP64/ILP64 and float/double variants It also tweaks some of the examples Change-Id: I8440c86be379839677bf65f5163e9d38cd17bb1f

Minor updates to FD framework

01c1d52

Also fixes nlls_example2.c bug Change-Id: Iec2ec4983eae41e229472d2bf902258a0a229d1d

Fix type for integer inform counters

f9e8978

Change-Id: I45d8f565b07c4c7330317061db53ed09941f5e3e

Add stand-alone entry-point to FD machinery

6451a2b

* jacobian_setup, * jacobian_calc and * jacobian_free * example to show how to call this framework instead of using the solver's internal mecanism Change-Id: I4eb24ce1cf67a78dceac541066f509657ef3d202

Add stand-alone FD Jacobian example

81e973d

Example to illustrate how to use RALFit FD machinery to estimate a Jacobian matrix outside of the solver Change-Id: I34b81558e6375c8973f6988a65f7708354f3b391

Fix bug is perturbation sign

a5adbfd

Update and tidy examples Change-Id: I094948f5705f836c7d2aa4905b6081ed6f0c6c06

Add FD support for weighted residuals

e1f4e2c

Added new extended examples Fixed bug in weight size passed in linesearch Change-Id: Ia58f6c8bf74e7da3999b2c69e48fed874c95d112

Add unit test

19fbf0c

* tweak examples * UT fd checker * UT fd solver Change-Id: I8f8d4fb8674f7058a453400b1d66449f0ea08c4d

Pretty print extremes for check derivative

73b2581

Change-Id: I223cb5906f8853b7b346bef5a7ef18f2304a772e

Reduce verbosity in derivative checker

3e0b06d

Change-Id: Icdbf3b11671b17cc9bb2248a3e60ba07d2042630

Remove ambiguous Public statement for derived type

e29fb04

Fixes ifort compilation error Change-Id: Ib6ece887276e4a2700c79d26d224182edf0d3ec4

Add Initial Fortran documentation

18f417d

Still need to move common chunks to `../common` Change-Id: Id611ab225488814649606de88bc3085c34f20689

Add C documentation

fb0a55d

Change-Id: Ibf48fa781dda37c4adc9cd581da0512dbc5177c3

update the docs for python 3 and rtd (#120)

4164ec8

* update the docs for python 3 and rtd * Fix warnings, including c issues * fix typos and c signature * Add a docs readme * Fix python warnings/errors

Minor fixes to C/Fortran documentation

286d4bc

* Typos * Removal of trailing blanks Change-Id: I85304c042b21728edfc544b12fa93b9a0d76eac7

Merge branch 'master' into fd

6e772a7

Add missing copyright banners

6e776fb

Change-Id: I2f3d96e9af7d878b40a54bc05197fc94a2fae070

as-amd added the enhancement label Aug 14, 2024

as-amd added this to the RC3: Enhancements to solver milestone Aug 14, 2024

as-amd requested a review from tyronerees August 14, 2024 12:58

as-amd assigned tyronerees and as-amd Aug 14, 2024

tyronerees requested changes Aug 30, 2024

View reviewed changes

as-amd and others added 5 commits September 18, 2024 11:27

Update libRALFit/doc/C/howtouse.rst

78b88e7

Suggestion accepted Co-authored-by: Tyrone Rees <[email protected]>

Update libRALFit/doc/C/howtouse.rst

ecfa21a

Co-authored-by: Tyrone Rees <[email protected]>

Update libRALFit/doc/C/howtouse.rst

ebdf67b

Co-authored-by: Tyrone Rees <[email protected]>

Update libRALFit/doc/C/howtouse.rst

4966a06

Co-authored-by: Tyrone Rees <[email protected]>

Update libRALFit/test/nlls_test.f90

8b1d820

Co-authored-by: Tyrone Rees <[email protected]>

as-amd and others added 8 commits September 18, 2024 11:29

Update libRALFit/test/nlls_test.f90

37fbbdc

Co-authored-by: Tyrone Rees <[email protected]>

Update libRALFit/test/nlls_test.f90

e99b209

Co-authored-by: Tyrone Rees <[email protected]>

Update libRALFit/test/nlls_test.f90

d172f2c

Co-authored-by: Tyrone Rees <[email protected]>

Update libRALFit/doc/C/howtouse.rst

c2f30ea

Co-authored-by: Tyrone Rees <[email protected]>

Updates based on peer review

9f074c2

Change-Id: Ic63a9aeacf759fc1e67c05955233b9b92f0a0b3f

Move derivative checker section to common

12ad570

Remove duplicated text and include from common file Change-Id: I984b3a7b33239bf282f5395be14d9691f186c790

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add Finite-Diffs (FD Branch) to RALFit #121

Add Finite-Diffs (FD Branch) to RALFit #121

as-amd commented Aug 14, 2024

tyronerees left a comment

tyronerees Aug 30, 2024

tyronerees Aug 30, 2024

as-amd Sep 2, 2024 •

edited

Loading

tyronerees Aug 30, 2024

as-amd Sep 18, 2024

tyronerees Aug 30, 2024

tyronerees Aug 30, 2024

as-amd Sep 2, 2024 •

edited

Loading

		@@ -202,6 +202,9 @@ It must have the following signature:

		:param status: \|eval_J_status\|

		If user does not provide the call-back it may be substituted by ``NULL`` and

		@@ -229,6 +232,8 @@ It must have the following signature:

		:param status: \|eval_Hf_status\|

		If user does not provide the call-back it may be substituted by ``NULL``.

Add Finite-Diffs (FD Branch) to RALFit #121

Are you sure you want to change the base?

Add Finite-Diffs (FD Branch) to RALFit #121

Conversation

as-amd commented Aug 14, 2024

This branch enables

tyronerees left a comment

Choose a reason for hiding this comment

tyronerees Aug 30, 2024

Choose a reason for hiding this comment

tyronerees Aug 30, 2024

Choose a reason for hiding this comment

as-amd Sep 2, 2024 • edited Loading

Choose a reason for hiding this comment

tyronerees Aug 30, 2024

Choose a reason for hiding this comment

as-amd Sep 18, 2024

Choose a reason for hiding this comment

tyronerees Aug 30, 2024

Choose a reason for hiding this comment

tyronerees Aug 30, 2024

Choose a reason for hiding this comment

as-amd Sep 2, 2024 • edited Loading

Choose a reason for hiding this comment

as-amd Sep 2, 2024 •

edited

Loading

as-amd Sep 2, 2024 •

edited

Loading