Skip to content

Latest commit

 

History

History
218 lines (167 loc) · 7.65 KB

to_PyG.md

File metadata and controls

218 lines (167 loc) · 7.65 KB

Convert Truss to Pytorch-Geometric HeteroData

Introduction

With the increasing importance of graph deep learning in the field of truss design, we also provide a solution to let our users convert the Truss object to the data structure of Pytorch-Geometric conveniently.
After slientruss3d v2.0.0, you can use slientruss3d.data.TrussHeteroDataCreator to build heterogeneous graph data (torch_geometric.data.HeteroData).

Installation

You have to install the Pytorch and PyTorch Geometric first:
Install Pytorch
Install PyTorch Geometric

How to use it ?

Constructor

TrussHeteroDataCreator(metapathType: MetapathType = MetapathType.NO_IMPLICIT, 
                       taskType    : TaskType     = TaskType.OPTIMIZATION   ) -> None

  • metapathType : Whether to use impicit connection of the bipartite graph representation of the truss.

    • MetapathType.USE_IMPLICIT
    • MetapathType.NO_IMPLICIT

  • taskType : Type of your task. It'll affect the fields of the HeteroData.

    • TaskType.OPTIMIZATION
    • TaskType.REGRESSION

( Both MetapathType and TaskType are in slientruss3d.type module. )


Build HeteroData from JSON file

TrussHeteroDataCreator.FromJSON(
    trussJSONFile, trussDim, 
    forceScale      = 1., 
    displaceScale   = 1., 
    positionScale   = 1., 
    usedMemberTypes = None, 
    fixedMemberType = MemberType(1., 1e7, 0.1), 
    isUseFixed      = True, 
    isOutputFile    = False
) -> torch_geometric.data.HeteroData
  • trussJSONFile : JSON file.
  • trussDim : Dimension of the truss.
  • forceScale : All the external force magnitudes in HeteroData will be divided by this value.
  • displaceScale : All the displacement magnitudes in HeteroData will be divided by this value.
  • positionScale : All the position ( x, y, (z) ) magnitudes in HeteroData will be divided by this value.
  • usedMemberTypes : If you want to do regression or imiation-learning task, please provide the list of used MemberTypes.
  • fixedMemberType : This parameter is used to get prior stress and displacement of the truss.
  • isUseFixed : Whether to include prior stress and displacement in HeteroData or not.
  • isOutputFile : Whether the JSON file is an output file or not.

Build HeteroData from Truss object

TrussHeteroDataCreator.FromTruss(
    truss, 
    forceScale      = 1., 
    displaceScale   = 1., 
    positionScale   = 1., 
    usedMemberTypes = None, 
    fixedMemberType = MemberType(1., 1e7, 0.1), 
    isUseFixed      = True, 
    trussSrc        = None
) -> torch_geometric.data.HeteroData
  • truss : Truss object.
  • forceScale : All the external force magnitudes in HeteroData will be divided by this value.
  • displaceScale : All the displacement magnitudes in HeteroData will be divided by this value.
  • positionScale : All the position ( x, y, (z) ) magnitudes in HeteroData will be divided by this value.
  • usedMemberTypes : If you want to do regression or imiation-learning task, please provide the list of used MemberTypes.
  • fixedMemberType : This parameter is used to get prior stress and displacement of the truss.
  • isUseFixed : Whether to include prior stress and displacement in HeteroData or not.
  • trussSrc : An identifier of the input truss. (Can be None, an integer, source JSON filename ... etc)

Add dense edges

TrussHeteroDataCreator.AddDenseEdges(graphData: torch_geometric.data.HeteroData) -> torch_geometric.data.HeteroData

Note that this method will modify the input data directly.

  • graphData : HeteroData.

Add master node

TrussHeteroDataCreator.AddMasterNode(graphData: torch_geometric.data.HeteroData, embeddingDim=1, fillValue=1.) -> torch_geometric.data.HeteroData

Note that this method will modify the input data directly.

  • graphData : HeteroData.
  • embeddingDim : Dimension of the initial master node embedding.
  • fillValue : Initial value of all the elements in master node embedding.

Some useful properties

  • Origin Truss object.
TrussHeteroDataCreator.truss : Truss
  • Source of the origin truss. You could use it to find the origin Truss object from a batch of HeteroData.
TrussHeteroDataCreator.source : Any
  • Mapping from the indices of HeteroData["joint"] to jointIDs in Truss object.
TrussHeteroDataCreator.jointIndexToID : list[int]
  • Mapping from the indices of HeteroData["member"] to memberIDs in Truss object.
TrussHeteroDataCreator.memberIndexToID : list[int]

Fields in HeteroData

For optimization task (TaskType.OPTIMIZATION)

  • X data :
    • Joint :
      • If isUseFixed is True :

        [ Position_X, Position_Y, Position_Z, Force_X, Force_Y, Force_Z, Prior_Displace_X, Prior_Displace_Y, Prior_Displace_Z, Is_Support ]

      • If isUseFixed is False :

        [ Position_X, Position_Y, Position_Z, Force_X, Force_Y, Force_Z, Is_Support ]

    • Member :
      • If isUseFixed is True :

        [ Centroid_X, Centroid_Y, Centroid_Z, sin(Angle_Z_Axis), cos(Angle_Z_Axis), sin(Angle_X_Axis), cos(Angle_X_Axis), Member_Length, Prior_Internal_Force ]

      • If isUseFixed is False :

        [ Centroid_X, Centroid_Y, Centroid_Z, sin(Angle_Z_Axis), cos(Angle_Z_Axis), sin(Angle_X_Axis), cos(Angle_X_Axis), Member_Length ]

  • Y data :
    • Joint :

      (No data)

    • Member :
      • If usedMemberTypes is not None :

        [Index_of_MemberType_in_usedMemberTypes]

      • If usedMemberTypes is None :

        (No data)

  • Other :
    • Origin weigth of each Truss :

      [ Origin_Weight ]

    • Source of each Truss :

      [ Source ]

For Regression task (TaskType.REGRESSION)

  • X data :
    • Joint :
      • If isUseFixed is True :

        [ Position_X, Position_Y, Position_Z, Force_X, Force_Y, Force_Z, Prior_Displace_X, Prior_Displace_Y, Prior_Displace_Z, Is_Support ]

      • If isUseFixed is False :

        [ Position_X, Position_Y, Position_Z, Force_X, Force_Y, Force_Z, Is_Support ]

    • Member :
      • If isUseFixed is True :

        [ Centroid_X, Centroid_Y, Centroid_Z, sin(Angle_Z_Axis), cos(Angle_Z_Axis), sin(Angle_X_Axis), cos(Angle_X_Axis), Member_Length, Prior_Axial_Force, Cross_Sectioal_Area ]

      • If isUseFixed is False :

        [ Centroid_X, Centroid_Y, Centroid_Z, sin(Angle_Z_Axis), cos(Angle_Z_Axis), sin(Angle_X_Axis), cos(Angle_X_Axis), Member_Length, Cross_Sectioal_Area ]

  • Y data :
    • Joint :

      [ Displace_X, Displace_Y, Displace_Z ]

    • Member :

      [ Axial_Stress ]

  • Other :
    • Origin weigth of each Truss :

      [ Origin_Weight ]

    • Source of each Truss :

      [ Source ]

Example code

The following is the example of TrussHeteroDataCreator:

from slientruss3d.data  import TrussHeteroDataCreator
from slientruss3d.type  import TaskType
from slientruss3d.truss import Truss 

JSON_FILE    = "./data/bar-25_input_0.json"
TRUSS_DIM    = 3
IS_USE_TRUSS = False

# Creator to create PyG's HeteroData:
creator = TrussHeteroDataCreator(taskType=TaskType.OPTIMIZATION)

# Create PyG's HeteroData from JSON file or Truss object:
if IS_USE_TRUSS:
    truss = Truss(TRUSS_DIM).LoadFromJSON(JSON_FILE)
    graph = creator.FromTruss(truss, trussSrc=JSON_FILE)
else:
    graph = creator.FromJSON(JSON_FILE, TRUSS_DIM)

# Print the structure of the HeteroData:
print("Hetero Graph Structure:\n" + "-" * 50)
print(graph)