Vectors for Replay Mode #9

richahert · 2020-02-26T09:50:17Z

This PR brings vector functionality to aste that is needed for the Replay Mode FSI tutorial.

The main changes are in partition_mesh.py and preciceMap.cpp.

# Conflicts: # CMakeLists.txt # src/common.cpp # src/common.hpp # src/eval_mesh.py # src/partition_mesh.py # src/preciceMap.cpp

…rsOn()

# Conflicts: # src/eval_mesh.py

fsimonis

The overall direction of the work looks good but the implementation needs some polish.

It is a bit unclear to me how to handle the user functions in the scripts. We need some more documentation, input sanitization and examples there.

The implementation in the preciceMap needs some refactoring. It can be simplified and reduced in many corners.
Important is that we are careful regarding the preCICE interface calls. ASTE is used for benchmarking some internals, so the runtime should not be overly affected by input sanitization.

Looking forward to your next request! 👍

fsimonis · 2020-04-02T11:05:49Z

src/partition_mesh.py

@@ -299,6 +297,7 @@ def parse_args():
    parser.add_argument("--log", "-l", dest="logging", default="INFO", 
            choices=["DEBUG", "INFO", "WARNING", "ERROR", "CRITICAL"], 
            help="Set the log level. Default is INFO")
+    parser.add_argument("--datadim", "-d", dest="datadim", default=1, type=int, help="Dimensions of the function. Default is 1 (Scalar function.")


At this point it is simply the dimension of the data.

fsimonis · 2020-04-02T11:15:15Z

src/eval_mesh.py

@@ -46,6 +47,8 @@ def parse_args():
    parser.add_argument("--log", "-l", dest="logging", default="INFO", 
            choices=["DEBUG", "INFO", "WARNING", "ERROR", "CRITICAL"], help="""Set the log level. 
            Default is INFO""")
+    parser.add_argument("--dimf","-d", dest="dimf", default="1", choices=["1","2", "3"],


This gets a bit tricky now as the function has to return a 1, 2, or 3-dimensional result.
Maybe we could add a verification step to check |f(v)| = dim. We could use the vertex coordinates of the first vertex in the mesh.

Those changes are leftovers from testing. For the tutorial we can omit that line.
Does anyone use this file except for the Bunny-Bloodcell example?
If you think it is very useful to extend eval_mesh.py to vectordata I can do that properly, otherwise I would discard the changes here.

fsimonis · 2020-04-02T11:22:53Z

src/mesh_io.py

+            if len(parts) == 4:
                pointdata.append(float(parts[3]))
+            if len(parts) == 6:
+                pointvector = ()
+                for i in range(3):
+                    pointvector += (float(parts[3+i]),)
+                pointdata.append(pointvector)


Could we simplify this by simply cutting the parts into the coordinate and the data on the position 3?

x y z u v w ^^^^^...... coordinate ^^^^^ data

I tried to simpyfy it, however we have to take care of 3 cases anyway: positions only, scalars where pointdata is a list of floats and vectors where pointdata is a list of lists.

fsimonis · 2020-04-02T11:23:52Z

src/mesh_io.py

@@ -121,17 +138,24 @@ def read_txt(filename):
    return points, cells, cell_types, pointdata


-def write_vtk(filename, points, cells = None, cell_types = None, pointdata = None, tag = None):
+def write_vtk(filename, points, cells = None, cell_types = None, pointdata = None, tag = "MyVectors", datadim=3):


Is there a reason why we need to set the tag to MyVectors by default?

fsimonis · 2020-04-02T11:26:58Z

src/mesh_io.py

+            if len(pointdata.shape) == 1 or pointdata.shape[1]==1: #scalar data
+                DataArray.InsertTuple1(i, pointdata[i])
+            elif len(pointdata.shape) > 1 and pointdata.shape[1]==2: #two dimensional data
+                DataArray.InsertTuple2(i, pointdata[i,0], pointdata[i,1])
+            elif len(pointdata.shape) > 1 and pointdata.shape[1]==3: #3D-data
+                DataArray.InsertTuple3(i, pointdata[i,0], pointdata[i,1], pointdata[i,2])


len(pointdata.shape) == 1 or pointdata.shape[1]==1

I would assume consistency here.
Maybe we can assert it after reading the file?

Actually pointdata.shape[1]==1 does not make sense for a 1D array.

fsimonis · 2020-04-02T12:15:46Z

src/precice_to_aste.py

+
+    for i in range(1,args.n+1):
+        files.append(args.SolverName +".dt" + str(i) + ".vtk")
+


Please check if the above all files actually exist and tell the user which ones are missing.
This makes debugging easier.

fsimonis · 2020-04-02T12:16:26Z

src/precice_to_aste.py

+                        help="""Name of the participant that exported the data. 
+                        Files are named SolverName.dti.vtk""")
+    parser.add_argument("--ntimesteps", "-n", dest="n", type= int, help="""Number of timesteps""")
+    parser.add_argument("--log", "-l", dest="logging", default="INFO", 


Are you actually logging anything? 😉

fsimonis · 2020-04-02T12:19:55Z

src/preciceMap.cpp

  std::vector<std::array<double, 3>> positions;
  std::vector<std::array<size_t, 2>> edges;
  std::vector<std::array<size_t, 3>> triangles;
  std::vector<double> data;
 };

+class VectorMesh : public Mesh{
+public:
+  VectorData data; //Overload data with Vector data


An alternative implementation could be to simply use std::vector<double> data with a size_t dataStride = 1 in Mesh. This way you just need to call writeBlockScalarData or writeBlockVectorData later.

This would save you to go through the trouble of adding an extra class and reduce code duplication.
Remember that this tool should be as minimal and lightweight as possible. (Someone needs to maintain it.)

fsimonis · 2020-04-02T12:25:25Z

src/preciceMap.cpp

 /// Reads the mesh from the file. If not read_data, zeros are returned for data.
-Mesh readMesh(const std::string& filename, bool read_data = true)
+Mesh readMesh(const std::string& filename, bool read_data = true, bool vectordata = false)


The flag can be deduced from the actual data.

fsimonis · 2020-04-02T12:31:34Z

src/preciceMap.cpp

+      size_t i=0;
+      VectorData initialdata = readVectorData(meshes[0]);
+
+      for(auto const& vertexID : vertexIDs){


As mentioned earlier, we should use the writeBlockVectorData variant. The non-block version is more expensive as it has to sanitize the user input for each call.

richahert · 2020-05-04T14:46:58Z

I reviewed the requested changes and hope that the pull request is ready to merge.
I opened the issue #10 because this PR only provides vector functionality for participant A and full vector support would be nice.

fsimonis · 2020-05-27T11:24:24Z

We have to put this PR on hold until we cleaned up the repository state after the latest changes.

davidscn · 2022-09-19T06:10:36Z

Superseded by #28

richahert and others added 27 commits November 4, 2019 10:48

adapt to new directory structure in precice

2dfd246

different prettyprint location

214e887

support writing vector data in a vtk

0b140e8

partition_mesh.py should work for vectors now

9c98322

vector functionality unfinished

73312f6

Merge remote-tracking branch 'upstream/master'

35ac49e

# Conflicts: # CMakeLists.txt # src/common.cpp # src/common.hpp # src/eval_mesh.py # src/partition_mesh.py # src/preciceMap.cpp

fixes after merging updates

1a2bdd4

3d vector support in preciceMap.cpp

c02aa36

support for precice export, still buggy

8d5e0c9

enable support to read from multiple vectors with reader.ReadAllVecto…

fb4f319

…rsOn()

add script for converting precice output

6e317fd

Merge branch 'master-upstream' into vectors

f122c4c

# Conflicts: # src/eval_mesh.py

add stuff to gitignore

0d09207

Delete bunny.vtk

8c45bc7

cleanup

395d2c5

timestep is implemented as precice time window

cfe67fd

use roundmesh in every round

f735d1b

add log to gitignore

243ee93

consistent dimensions for data

d8699ca

cleanup

1b24eb8

cleanup

f73d999

cleanup

069f941

further cleanup

cd79817

github wants a new line at end of file

062ee1e

revert to scalar case

897d13c

write Vectordata with --vectordata flag

f03cd29

consistent use of datadim for dimension of data vectors.

0b2258d

BenjaminRodenberg self-requested a review March 11, 2020 08:27

uekerman requested review from fsimonis and removed request for BenjaminRodenberg March 12, 2020 14:21

fsimonis requested changes Apr 2, 2020

View reviewed changes

richahert added 16 commits April 15, 2020 13:42

some requested changes

33dce04

Remove subprocesses and use mesh_io functions directly

948996f

remove leftovers

4bbf42d

its a tag for data, not a function

428fc10

refactor write process for vectors

252ab7c

remove --vectordata tag

94537d4

delete leftovers

53c6f88

remove pointless or statement

b81387b

simplify read_txt

1936e8b

remove unnecessary imports

d124d0f

delete vector_demo.sh

13f1c56

revert chagens

944e457

delete readVectorData() because general function exists

9b0423e

remove vectordata tag because it is automated

f7c27d1

just formatting

fdedc9e

formatting againg..

84ebcbc

richahert mentioned this pull request May 4, 2020

Vectors for Participant B #10

Closed

richahert changed the title ~~WIP: Vectors for Replay Mode~~ Vectors for Replay Mode May 4, 2020

BenjaminRodenberg mentioned this pull request May 8, 2020

WIP: aste Replay Mode tutorial on elastic flap case precice/tutorials#73

Open

BenjaminRodenberg requested a review from fsimonis May 11, 2020 14:21

fsimonis marked this pull request as draft May 27, 2020 11:24

BenjaminRodenberg mentioned this pull request Jul 21, 2020

Create Tutorial for "Replay Mode" precice/tutorials#45

Open

BenjaminRodenberg self-assigned this Jul 21, 2020

fsimonis mentioned this pull request Apr 30, 2021

Fully port ASTE to vtk #21

Closed

davidscn closed this Sep 19, 2022

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Vectors for Replay Mode #9

Vectors for Replay Mode #9

richahert commented Feb 26, 2020

fsimonis left a comment

fsimonis Apr 2, 2020

fsimonis Apr 2, 2020

richahert Apr 15, 2020

fsimonis Apr 2, 2020

richahert May 4, 2020

fsimonis Apr 2, 2020

fsimonis Apr 2, 2020

richahert May 4, 2020

fsimonis Apr 2, 2020

fsimonis Apr 2, 2020

fsimonis Apr 2, 2020

fsimonis Apr 2, 2020

fsimonis Apr 2, 2020

richahert commented May 4, 2020

fsimonis commented May 27, 2020

davidscn commented Sep 19, 2022


		for i in range(1,args.n+1):
		files.append(args.SolverName +".dt" + str(i) + ".vtk")

Vectors for Replay Mode #9

Vectors for Replay Mode #9

Conversation

richahert commented Feb 26, 2020

fsimonis left a 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

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

richahert commented May 4, 2020

fsimonis commented May 27, 2020

davidscn commented Sep 19, 2022