In C++ I would just compare the memory addresses of both objects. How would I do something similar in MATLAB?
Worst Case would be to have a static variable that iterates in each constructor and every object gets the current value as ID. But is there a better solution?
Thank you in advance.
#Edit:
I'd like to extend this question by assuming I have some given/not changeable classes inheriting handle and overloading eq. If I want to compare two objects of this class can I somehow cast both instances to handle and use the implementation of eq of the super class?
To test that two handle objects a and b refer to the same instance, you only need to use a == b. This is the same as eq(a, b). This is the defined behaviour of == for handle objects. I.e., for handle objects, == tests for equality of instances, not equality of the values within the instances. This is different from value objects.
For this to work you need to be using handle objects (classdef myObject < handle) because it doesn't make sense to test instances of value objects.
N.B. if you also need to get some kind of instance identifier for a handle object, then you need to do something like you describe using a persistent variable. Here's an example. In that case I would make that a base class for all your objects, so you wouldn't have to copy the same code into each class. But that's unnecessary if all you want to do is test two instances.
Related
Consider the following toy class:
class myGiantClass(){
val serializableElement = ...
// lots of other variables and methods here
}
// main program
val listOfGiantObjects: List[myGiantClass] = ....
What I need is to serialize/deserialize listOfGiantObjects. The issue is that myGiantClass contains lots of junk objects and variables which I don't/can't serialize/deserialize. Instead the only element of the myGiantClass that I want to serialize is serializableElement inside each object of listOfGiantObjects.
So after deserialize, listOfGiantObjects is expected to contain a bunch of myGiantClass objects which contain only serializableElement (the rest set to default).
Any ideas?
Of course there are two approaches (or defaults): all elements should be serialized by default, or none.
Within the "all" scenario, you could take a look at the #transient annotation, for marking fields that should not be serialized.
It may seem an unoptimal approach in case of a large number of elements that should not be serialized. However, it does communicate what you are trying to achieve. Moreover, you could arrange your code using composition or inner classes to better define the scope of serialization.
At last resort, ad-hoc serializaion with custom attributes is a way (e.g., to implement the none-by-default scenario).
I want to change the format of my data, from RDD(Label:String,(ID:String,Data:Array[Double])) to an RDD Object with the label, id and data as components.
But when I print my RDD consecutively twice, the references of objects change :
class Data_Object(private val id:String, private var vector:Vector) extends Serializable {
var label = ""
...
}
First print
(1,ms3.Data_Object#35062c11)
(2,ms3.Data_Object#25789aa9)
Second print
(2,ms3.Data_Object#6bf5d886)
(1,ms3.Data_Object#a4eb65)
I think that explains why the subtract method doesn't work. So can I use subtract with objects as values, or do I return to my classic model ?
Unless you specify otherwise, objects in Scala (and Java) are compared using reference equality (i.e. their memory address). They are also printed out according to this address, hence the Data_Object#6bf5d886 and so on.
Using reference equality means that two Data_Object instances with identical properties will NOT compare as equal unless they are exactly the same object. Also, their references will change from one run to the next.
Particularly in a distributed system like Spark, this is no good - we need to be able to tell whether two objects in two different JVMs are the same or not, according to their properties. Until this is fixed, RDD operations like subtract will not give the results you expect.
Fortunately, this is usually easy to fix in Scala/Spark - define your class as a case class. This automatically generates equals and hashcode and toString methods derived from all of the properties of the class. For example:
case class Data_Object(id:String, label:String, vector:Vector)
If you want to compare your objects according to only some of the properties, you'll have to define your own equals and hashcode methods, though. See Programming in Scala, for example.
I am having trouble understanding with some of the code snippets about this part of the Java tutorial: http://docs.oracle.com/javase/tutorial/java/IandI/interfaceAsType.html
public Object findLargest(Object object1, Object object2) {
Relatable obj1 = (Relatable)object1;
Relatable obj2 = (Relatable)object2;
if ((obj1).isLargerThan(obj2) > 0)
return object1;
else
return object2;
}
and:
public interface Relatable {
// this (object calling isLargerThan)
// and other must be instances of
// the same class returns 1, 0, -1
// if this is greater than,
// equal to, or less than other
public int isLargerThan(Relatable other);
}
In the first example, why am I downcasting Object types into Relatable types? What happens if the first method doesn't include the first two statements?
Let's say I wrote a Rectangle class that implements the Relatable interface and has the "findLargest" method. If I know that I'm comparing two Rectangle objects, why not just make the first method downcast the objects into Rectangles instead?
You cast the Objects into Relatable types because otherwise you cannot use the methods declared in the Relatable interface. Since Object does not have the isLargerThan method, you would get a compiler error without casting. Honestly, in my opinion the findLargest method as shown here was not very well designed; a better illustration of the purpose of Interfaces would be to ask for Relatable objects as the parameters like so:
public Object findLargest(Relatable object1, Relatable object2) {
//implementation not shown to save space
}
This way, the user must pass Relatable objects, but they can pass any object whose class implements Relatable (such as Rectangle)
"If I know that I'm comparing two Rectangle objects..."True, if you know that you are comparing two Rectangle objects, there is little use for an interface, but the purpose of interfaces is to allow you to create a generic "type" of object that can be used to define common features of several different classes.For example, what if you also had a Circle class and a Square class (both of which implemented Relatable)? In this case, you do not necessarily know the exact type of object you have, but you would know that it is Relatable, so it would be best to cast to type Relatable and use the isLargerThan method in a case like this.
Interfaces define a set of methods which every class which the interface implements has to implement. The downcast is necessary to get access to these methods.
You don't know if you are comparing rectangles with this interface. You could get any Relatble passed. This is one of the cases generics come in handy.
1.In the first example, why am I down casting Object types into Relatable types? What happens if the first method doesn't include the first two statements?
Answer
Every object has some basic functionality and you want a specific object write now. You are down casting your object into a "Relatable" so you can use the "isLargerThan" method(an object wont have it since it has only basic common stuff).
If you didn't down cast, you would not pass compilation.
2.Let's say I wrote a Rectangle class that implements the Relatable interface and has the "findLargest" method. If I know that I'm comparing two Rectangle objects, why not just make the first method downcast the objects into Rectangles instead?
Answer
Since you want to create something generic.
Lets say you have a Student and a Driver. Both of them are People. You can create an interface called IPeople and make both the Student and the driver implement it.
IPeople will have a method called "getAge()" that each of them will implement.
IPeople will have all the functionality that you need for "People". That's how you create cross object functionality under the "same hat".
Hi I'm using an object with superclass dynamicprops & matlab.mixin.Copyable
Now I would like to now the size (memorywise) of each of the dynamic properties. I tried stuff like whos and getfield. But it seems like I have a very hard time to find something.
I know that people strugle to get exactly this for handle objects, which this is one of them... But I was wondering if somebody has a solution on that nevertheless.
yours
magu_
You could use the properties function to get a list of properties exposed by an object. The same can be done using metaclass. You can then iterate over them and use x.(p) syntax to access each by name (dynamic field names). You would determine the size memory-wise the same way you do any other variable (size/class or whos).
Note that you might need to recursively traverse the properties in case they are containers themselves (objects, structs, cell-arrays).
This question already has answers here:
Closed 10 years ago.
Possible Duplicate:
Why does fast enumeration not skip the NSNumbers when I specify NSStrings?
I noticed some unexpected behavior while using fast enumeration recently. In hindsight I was probably expecting fast enumeration to do more than it's intended for, so I'm looking for some clarification on how it's actually works.
Say I have a parent class Shape with 2 child classes, 3SidedShape and 4SidedShape. I have an array called myShapes, that contains objects from both the 3 and 4 sided classes.
If I wanted to search through the array myShapes, but I'm only concerned with 3 sides shapes what I was doing is:
for (3SidedShape *shape in myShapes)
My thought was that I would only be iterating over objects of class 3SidedShape, but that is not the case? I guess I'm casting all objects as 3SidedShape whether they like it or not. I'm evening returning the object after as a completely different class. Granted I'm not calling any methods that both classes don't have, but I didn't expect class siblings to just re-cast so easily without a hitch? Did I just get lucky here or can you really enumerate as any class you please regardless of relation? Can anyone explain what actually happens during enumeration?
The type specified in a for...in loop, aka fast enumeration, casts all the elements in the collection to the specified type. The reason why they are "re-cast so easily" is that casting does NOT turn one type of object into another (how would that work?). It's a hint to the compiler telling it to treat the object as if it were the other type, as if to say "don't worry, this object is of (insert type), so type check it as such." Sending the object a message it can't handle, but the type it was casted to can, will still crash the app. What you should do is this:
for (id shape in myShapes){
if ([shape isKindOfClass: [3SidedShape class]]){
//insert code here
}
}
That code assumes nothing of type, using introspection to only perform the code for objects who are of type 3SidedShape or a subclass of 3SidedShape. For exact checking (excluding subclasses) use isMemberOfClass:. Be wary of using isMemberOfClass: to test membership of a class in a class cluster (NSNumber), however, due to their more complex internal implementation.