Package sim.field.continuous

Class Continuous3D

java.lang.Object
sim.field.SparseField
sim.field.continuous.Continuous3D
All Implemented Interfaces:
Serializable, SparseField3D
public class Continuous3D extends SparseField implements SparseField3D
A storage facility for objects located in a continuous 3D environment. This facility relates objects with 3D double tuples (in the form of Double3D). The facility extends SparseField, and like other objects which extend SparseField (such as SparseGrid3D), the facility can only relate any given object with a single location at a time -- that is, an object cannot hold two locations at once in the Continuous3D field.

Because hashtable lookups are more expensive than just storing the object, we suggest that you ALSO store the location of an object in the object itself, so you can read from the object rather than having to call getObjectLocation(object).

The Continuous3D has been arranged to make neighborhood lookup information efficient. It discretizes the space into grid buckets. The discretization size of the buckets is provided in the constructor and cannot be changed thereafter. If the discretization was 0.7, for example, then one bucket would be (0,0,0) to (under 0.7, under 0.7, under 0.7), another bucket would be (0,0,0.0,0.7) to (under 0.7, under 0.7, under 1.4), etc.

You can use Continuous3D to look up objects in a given region by asking for objects within the enclosing buckets, then rummaging through the buckets to find the individuals actually in the desired region. The trick here is to come up with a good bucket size. If the bucket size is much larger than the typical size of a neighborhood lookup, then a typical lookup will include large numbers of objects you don't care about; in the worst case, this is an O(n) lookup for something that could have been much smaller. On the other hand, if the bucket size is much smaller than the typical size of a neighborhood lookup, then you have to do lots of bucket lookups to cover your range; many if not most of these buckets could be empty. This can also be highly inefficient.

If your objects are point objects, you have no minimum bound on the discretization size. But if the object are non-point location objects (that is, they have dimensions of width, height, etc.), and you care about this overlap when you do distance lookups, then you have a minimum bound on your discretization. In this case, you want to make certain that your discretization is at LEAST larger than the LARGEST dimension of any object you plan on putting in the Continuous3D. The idea here is that if an any part of an object fell within the bounding box for your distance lookup task (see getNeighborsWithinDistance(...)), you're guaranteed that the stored location of the object must be within a bounding box 1 discretization larger in each direction.

Okay, so that gives you the minimum discretization you should use. What about the maximum discretization? It depends largely on the number of objects expected to occupy a given discretized bucket region, and on what kind of lookups you need to do for objects within a given distance. Searching through one bucket is a hash table lookup. A smaller discretization returns a more accurate sample of objects within the requested bounding box, but requires more hash table lookups. If you have point location objects, and your field is very dense (LOTS of objects in a bucket on average), then we recommend a discretization equal to the maximum range distance you are likely to look up; but if your field is very sparse, then we recommend a discretization equal to twice the maximum range distance. You have to tune it. If you have non-point-location objects, then you have two choices. One approach is to assume a discretization equal to the maximum range distance, but when doing lookups with getNeighborsWithinDistance(...), you need to state that you're using non-point-location objects. If you're fairly sparse and your objects aren't big, you can set the discretization to twice the maximum range distance, and you should be safe calling getNeighborsWithinDistance() pretending that your objects are point-location; this saves you a lot of hash table lookups.

At any rate, do NOT go below the minimum discretization rules.

But wait, you say, I have objects of widely varying sizes. Or I have many different neighborhood lookup range needs. Never fear. Just use multiple Continuous3Ds of different discretizations. Depending on your needs, you can put all the objects in all of the Continuous3Ds (making different range lookups efficient) or various-sized classes of objects in their own Continuous3Ds perhaps. You have to think this through based on your needs. If all the objects were in all of the Continuous3Ds, you'd think that'd be inefficient in moving objects around. Not really: if the discretizations doubled (or more) each time, you're looking at typically an O(ln n) number of Continuous3Ds, and a corresponding number of lookups.

Continuous3D objects have a width and a height, but this is only used in computing toroidal (wrap-around) situations. If you don't care about toroidal features, then you can completely disregard the width and height.

See Also:

  • Field Details

    • doubleLocationHash

      public Map doubleLocationHash
      Where we store the Double3D values hashed by object

    • width

      public double width

    • height

      public double height

    • length

      public double length

    • discretization

      public final double discretization

  • Constructor Details

    • Continuous3D

      public Continuous3D(double discretization, double width, double height, double length)
      Provide expected bounds on the SparseContinuous3D

    • Continuous3D

      public Continuous3D(Continuous3D other)

  • Method Details

    • getObjectLocation

      public final Double3D getObjectLocation(Object obj)

    • discretize

      public final Int3D discretize(Double3D location)
      Discretizes the location according to the internal discretization of the Continuous3D. You can use this to determine what internal grid slot the continuous point would fall in.

    • discretize

      public final Int3D discretize(Double3D location, int discretization)
      Discretizes the location according to the provided discretization, which may or may not be the discretization used internally by the Continuous3D. If you're trying to determine what grid slot a continuous point would fall in, you probably want discretize(location) instead.

    • setObjectLocation

      public final boolean setObjectLocation(Object obj, Double3D location)

    • clear

      public final Bag clear()
      Description copied from class: SparseField
      Deletes everything, returning all the objects as a Bag (which you can freely use and modify). If you need the Bag, then this is a useful method -- otherwise it might in fact be faster to just make a brand new Sparse Field and let the garbage collector do its magic.
      Overrides:
      clear in class SparseField

    • remove

      public final Object remove(Object obj)
      Description copied from class: SparseField
      Removes an object if it exists. Returns its location, or null if the object didn't exist.
      Overrides:
      remove in class SparseField

    • getWidth

      public double getWidth()
      Get the width

    • getHeight

      public double getHeight()
      Get the height

    • getLength

      public double getLength()
      Get the height

    • tx

      public final double tx(double x)
      Toroidal x

    • ty

      public final double ty(double y)
      Toroidal y

    • tz

      public final double tz(double z)
      Toroidal z

    • stx

      public double stx(double x)
      Simple [and fast] toroidal x. Use this if the values you'd pass in never stray beyond (-width ... width * 2) not inclusive. It's a bit faster than the full toroidal computation as it uses if statements rather than two modulos. The following definition:
      { double width = this.width; if (x >= 0) { if (x invalid input: '<' width) return x; return x - width; } return x + width; }
      ...produces the shortest code (24 bytes) and is inlined in Hotspot for 1.4.1. However removing the double width = this.width; is likely to be a little faster if most objects are within the toroidal region.

    • sty

      public double sty(double y)
      Simple [and fast] toroidal y. Use this if the values you'd pass in never stray beyond (-height ... height * 2) not inclusive. It's a bit faster than the full toroidal computation as it uses if statements rather than two modulos. The following definition:
      { double height = this.height; if (y >= 0) { if (y invalid input: '<' height) return y ; return y - height; } return y + height; }
      ...produces the shortest code (24 bytes) and is inlined in Hotspot for 1.4.1. However removing the double height = this.height; is likely to be a little faster if most objects are within the toroidal region.

    • stz

      public double stz(double z)
      Simple [and fast] toroidal z. Use this if the values you'd pass in never stray beyond (-length ... length * 2) not inclusive. It's a bit faster than the full toroidal computation as it uses if statements rather than two modulos. The following definition:
      { double length = this.length; if (z >= 0) { if (z invalid input: '<' length) return z ; return z - length; } return z + length; }
      ...produces the shortest code (24 bytes) and is inlined in Hotspot for 1.4.1. However removing the double length = this.length; is likely to be a little faster if most objects are within the toroidal region.

    • tdx

      public double tdx(double x1, double x2)
      Minimum toroidal difference between two values in the X dimension.

    • tdy

      public double tdy(double y1, double y2)
      Minimum toroidal difference between two values in the Y dimension.

    • tdz

      public double tdz(double z1, double z2)
      Minimum toroidal difference between two values in the Z dimension.

    • tds

      public double tds(Double3D d1, Double3D d2)
      Minimum Toroidal Distance Squared between two points. This computes the "shortest" (squared) distance between two points, considering wrap-around possibilities as well.

    • tv

      public Double3D tv(Double3D d1, Double3D d2)
      Minimum Toroidal difference vector between two points. This subtracts the second point from the first and produces the minimum-length such subtractive vector, considering wrap-around possibilities as well

    • getObjectsExactlyWithinDistance

      public Bag getObjectsExactlyWithinDistance(Double3D position, double distance)
      Deprecated.
      Returns a Bag containing EXACTLY those objects within a certain distance of a given position, or equal to that distance, measuring using a circle of radius 'distance' around the given position. Assumes non-toroidal point objects.

      Note: if the field is toroidal, and position is outside the boundaries, it will be wrapped to within the boundaries before computation.

    • getObjectsExactlyWithinDistance

      public Bag getObjectsExactlyWithinDistance(Double3D position, double distance, boolean toroidal)
      Deprecated.
      Returns a Bag containing EXACTLY those objects within a certain distance of a given position, or equal to that distance, measuring using a circle of radius 'distance' around the given position. If 'toroidal' is true, then the distance is measured assuming the environment is toroidal. Assumes point objects.

      Note: if the field is toroidal, and position is outside the boundaries, it will be wrapped to within the boundaries before computation.

    • getObjectsExactlyWithinDistance

      public Bag getObjectsExactlyWithinDistance(Double3D position, double distance, boolean toroidal, boolean radial, boolean inclusive, Bag result)
      Deprecated.
      Returns a Bag containing EXACTLY those objects within a certain distance of a given position. If 'radial' is true, then the distance is measured using a circle around the position, else the distance is meaured using a square around the position (that is, it's the maximum of the x and y distances). If 'inclusive' is true, then objects that are exactly the given distance away are included as well, else they are discarded. If 'toroidal' is true, then the distance is measured assuming the environment is toroidal. If the Bag 'result' is provided, it will be cleared and objects placed in it and it will be returned, else if it is null, then this method will create a new Bag and use that instead. Assumes point objects.

      Note: if the field is toroidal, and position is outside the boundaries, it will be wrapped to within the boundaries before computation.

    • getNeighborsExactlyWithinDistance

      public Bag getNeighborsExactlyWithinDistance(Double3D position, double distance)
      Returns a Bag containing EXACTLY those objects within a certain distance of a given position, or equal to that distance, measuring using a circle of radius 'distance' around the given position. Assumes non-toroidal point objects.

      Note: if the field is toroidal, and position is outside the boundaries, it will be wrapped to within the boundaries before computation.

    • getNeighborsExactlyWithinDistance

      public Bag getNeighborsExactlyWithinDistance(Double3D position, double distance, boolean toroidal)
      Returns a Bag containing EXACTLY those objects within a certain distance of a given position, or equal to that distance, measuring using a circle of radius 'distance' around the given position. If 'toroidal' is true, then the distance is measured assuming the environment is toroidal. Assumes point objects.

      Note: if the field is toroidal, and position is outside the boundaries, it will be wrapped to within the boundaries before computation.

    • getNeighborsExactlyWithinDistance

      public Bag getNeighborsExactlyWithinDistance(Double3D position, double distance, boolean toroidal, boolean radial, boolean inclusive, Bag result)
      Returns a Bag containing EXACTLY those objects within a certain distance of a given position. If 'radial' is true, then the distance is measured using a circle around the position, else the distance is meaured using a square around the position (that is, it's the maximum of the x and y distances). If 'inclusive' is true, then objects that are exactly the given distance away are included as well, else they are discarded. If 'toroidal' is true, then the distance is measured assuming the environment is toroidal. If the Bag 'result' is provided, it will be cleared and objects placed in it and it will be returned, else if it is null, then this method will create a new Bag and use that instead. Assumes point objects.

      Note: if the field is toroidal, and position is outside the boundaries, it will be wrapped to within the boundaries before computation.

    • getObjectsWithinDistance

      public Bag getObjectsWithinDistance(Double3D position, double distance)
      Deprecated.
      Returns a bag containing AT LEAST those objects within the bounding box surrounding the specified distance of the specified position. The bag could include other objects than this. In this case we include the object if any part of the bounding box could overlap into the desired region. To do this, if nonPointObjects is true, we extend the search space by one extra discretization in all directions. For small distances within a single bucket, this returns 27 bucket's worth rather than 1 (!!!), so if you know you only care about the actual x/y/z points stored, rather than possible object overlap into the distance sphere you specified, you'd want to set nonPointObjects to FALSE. That's a lot of extra hash table lookups. [assumes non-toroidal, point objects]

      Note: if the field is toroidal, and position is outside the boundaries, it will be wrapped to within the boundaries before computation.

    • getObjectsWithinDistance

      public Bag getObjectsWithinDistance(Double3D position, double distance, boolean toroidal)
      Deprecated.
      Returns a bag containing AT LEAST those objects within the bounding box surrounding the specified distance of the specified position. The bag could include other objects than this. If toroidal, then wrap-around possibilities are also considered. In this case we include the object if any part of the bounding box could overlap into the desired region. To do this, if nonPointObjects is true, we extend the search space by one extra discretization in all directions. For small distances within a single bucket, this returns 27 bucket's worth rather than 1 (!!!), so if you know you only care about the actual x/y/z points stored, rather than possible object overlap into the distance sphere you specified, you'd want to set nonPointObjects to FALSE. That's a lot of extra hash table lookups. [assumes point objects]

      Note: if the field is toroidal, and position is outside the boundaries, it will be wrapped to within the boundaries before computation.

    • getObjectsWithinDistance

      public Bag getObjectsWithinDistance(Double3D position, double distance, boolean toroidal, boolean nonPointObjects)
      Deprecated.
      Returns a bag containing AT LEAST those objects within the bounding box surrounding the specified distance of the specified position. The bag could include other objects than this. If toroidal, then wrap-around possibilities are also considered. If nonPointObjects, then it is presumed that the object isn't just a point in space, but in fact fills an area in space where the x/y point location could be at the extreme corner of a bounding box of the object. In this case we include the object if any part of the bounding box could overlap into the desired region. To do this, if nonPointObjects is true, we extend the search space by one extra discretization in all directions. For small distances within a single bucket, this returns 27 bucket's worth rather than 1 (!!!), so if you know you only care about the actual x/y/z points stored, rather than possible object overlap into the distance sphere you specified, you'd want to set nonPointObjects to FALSE. That's a lot of extra hash table lookups.

      Note: if the field is toroidal, and position is outside the boundaries, it will be wrapped to within the boundaries before computation.

    • getObjectsWithinDistance

      public Bag getObjectsWithinDistance(Double3D position, double distance, boolean toroidal, boolean nonPointObjects, Bag result)
      Deprecated.
      Puts into the result Bag (and returns it) AT LEAST those objects within the bounding box surrounding the specified distance of the specified position. The bag could include other objects than this. If toroidal, then wrap-around possibilities are also considered. If nonPointObjects, then it is presumed that the object isn't just a point in space, but in fact fills an area in space where the x/y point location could be at the extreme corner of a bounding box of the object. In this case we include the object if any part of the bounding box could overlap into the desired region. To do this, if nonPointObjects is true, we extend the search space by one extra discretization in all directions. For small distances within a single bucket, this returns 27 bucket's worth rather than 1 (!!!), so if you know you only care about the actual x/y/z points stored, rather than possible object overlap into the distance sphere you specified, you'd want to set nonPointObjects to FALSE. That's a lot of extra hash table lookups.

      Note: if the field is toroidal, and position is outside the boundaries, it will be wrapped to within the boundaries before computation.

    • getNeighborsWithinDistance

      public Bag getNeighborsWithinDistance(Double3D position, double distance)
      Returns a bag containing AT LEAST those objects within the bounding box surrounding the specified distance of the specified position. The bag could include other objects than this. In this case we include the object if any part of the bounding box could overlap into the desired region. To do this, if nonPointObjects is true, we extend the search space by one extra discretization in all directions. For small distances within a single bucket, this returns 27 bucket's worth rather than 1 (!!!), so if you know you only care about the actual x/y/z points stored, rather than possible object overlap into the distance sphere you specified, you'd want to set nonPointObjects to FALSE. That's a lot of extra hash table lookups. [assumes non-toroidal, point objects]

      Note: if the field is toroidal, and position is outside the boundaries, it will be wrapped to within the boundaries before computation.

    • getNeighborsWithinDistance

      public Bag getNeighborsWithinDistance(Double3D position, double distance, boolean toroidal)
      Returns a bag containing AT LEAST those objects within the bounding box surrounding the specified distance of the specified position. The bag could include other objects than this. If toroidal, then wrap-around possibilities are also considered. In this case we include the object if any part of the bounding box could overlap into the desired region. To do this, if nonPointObjects is true, we extend the search space by one extra discretization in all directions. For small distances within a single bucket, this returns 27 bucket's worth rather than 1 (!!!), so if you know you only care about the actual x/y/z points stored, rather than possible object overlap into the distance sphere you specified, you'd want to set nonPointObjects to FALSE. That's a lot of extra hash table lookups. [assumes point objects]

      Note: if the field is toroidal, and position is outside the boundaries, it will be wrapped to within the boundaries before computation.

    • getNeighborsWithinDistance

      public Bag getNeighborsWithinDistance(Double3D position, double distance, boolean toroidal, boolean nonPointObjects)
      Returns a bag containing AT LEAST those objects within the bounding box surrounding the specified distance of the specified position. The bag could include other objects than this. If toroidal, then wrap-around possibilities are also considered. If nonPointObjects, then it is presumed that the object isn't just a point in space, but in fact fills an area in space where the x/y point location could be at the extreme corner of a bounding box of the object. In this case we include the object if any part of the bounding box could overlap into the desired region. To do this, if nonPointObjects is true, we extend the search space by one extra discretization in all directions. For small distances within a single bucket, this returns 27 bucket's worth rather than 1 (!!!), so if you know you only care about the actual x/y/z points stored, rather than possible object overlap into the distance sphere you specified, you'd want to set nonPointObjects to FALSE. That's a lot of extra hash table lookups.

      Note: if the field is toroidal, and position is outside the boundaries, it will be wrapped to within the boundaries before computation.

    • getNeighborsWithinDistance

      public Bag getNeighborsWithinDistance(Double3D position, double distance, boolean toroidal, boolean nonPointObjects, Bag result)
      Puts into the result Bag (and returns it) AT LEAST those objects within the bounding box surrounding the specified distance of the specified position. The bag could include other objects than this. If toroidal, then wrap-around possibilities are also considered. If nonPointObjects, then it is presumed that the object isn't just a point in space, but in fact fills an area in space where the x/y point location could be at the extreme corner of a bounding box of the object. In this case we include the object if any part of the bounding box could overlap into the desired region. To do this, if nonPointObjects is true, we extend the search space by one extra discretization in all directions. For small distances within a single bucket, this returns 27 bucket's worth rather than 1 (!!!), so if you know you only care about the actual x/y/z points stored, rather than possible object overlap into the distance sphere you specified, you'd want to set nonPointObjects to FALSE. That's a lot of extra hash table lookups.

      Note: if the field is toroidal, and position is outside the boundaries, it will be wrapped to within the boundaries before computation.

    • getObjectsAtDiscretizedLocation

      public Bag getObjectsAtDiscretizedLocation(Int3D location)
      Returns a bag containing all the objects at a given discretized location, or null when there are no objects at the location. You should NOT MODIFY THIS BAG. This is the actual container bag, and modifying it will almost certainly break the Sparse Field object. If you want to modify the bag, make a copy and modify the copy instead, using something along the lines of new Bag(foo.getObjectsAtLocation(location)) . Furthermore, changing values in the Sparse Field may result in a different bag being used -- so you should not rely on this bag staying valid. The default implementation of this method simply calls getRawObjectsAtLocation(), but you may need to override it for more custom functionality (which is rare).

    • getObjectsAtLocation

      public Bag getObjectsAtLocation(Double3D location)
      Returns a bag containing all the objects at a given location, or null if there are no such objects or if location is null. Unlike other SparseField versions, you may modify this bag.

    • numObjectsAtLocation

      public int numObjectsAtLocation(Double3D location)
      Returns the number of the objects at a given location, or 0 if there are no such objects or if location is null.

    • getObjectsAtLocationOfObject

      public Bag getObjectsAtLocationOfObject(Object obj)
      Returns a bag containing all the objects at the exact same location as a given object, including the object itself, or null if the object is not in the Field. Unlike other SparseField versions, you may modify this bag.
      Overrides:
      getObjectsAtLocationOfObject in class SparseField

    • numObjectsAtLocationOfObject

      public int numObjectsAtLocationOfObject(Object obj)
      Returns the number of objects at the exact same location as a given object, including the object itself, or 0 if the object is not in the Field.
      Overrides:
      numObjectsAtLocationOfObject in class SparseField

    • removeObjectsAtLocation

      public Bag removeObjectsAtLocation(Double3D location)
      Removes objects at exactly the given location, and returns a bag of them, or null of no objects are at that location. The Bag may be empty, or null, if there were no objects at that location. You can freely modify this bag.

    • getDimensions

      public final Double3D getDimensions()
      Description copied from interface: SparseField3D
      Returns the width and height of the sparse field as a Double3D
      Specified by:
      getDimensions in interface SparseField3D

    • getObjectLocationAsDouble3D

      public Double3D getObjectLocationAsDouble3D(Object obj)
      Returns the object location as a Double3D, or as null if there is no such object.
      Specified by:
      getObjectLocationAsDouble3D in interface SparseField3D