On Nov 23, 2007 4:43 PM, Sebastian Haase <haase@msg.ucsf.edu> wrote:

On Nov 23, 2007 3:37 PM, Pierre GM <pgmdevlist@gmail.com> wrote:

...

On Friday 23 November 2007 03:25:37 Sebastian Haase wrote:

...

Hi, this question might habe been answered before: I have my own ndarray-derived class. I did this, so that I can add another "custom attribute" -- let's say arr.filename

Sebastian, Could you post the __new__ and __array_finalize__ for your class ? Are you sure you update the attribute in __array_finalize__ ? And this page may come handy: http://www.scipy.org/Subclasses

Aah - I guess I did not do my homework ;-) Here is my code: class ndarray_inMrcFile(N.memmap): pass

data = self.data data.__class__ = ndarray_inMrcFile

Two points: 1.) Please don't get distracted by the way how I change the class type "after the fact". I get the data as a memmaped slice from a file. Then I do some (potentially not entirely clean) acrobatic to have this slice change it's class-type so that I can attach the original memmap (+ other attributes, such as filename) onto the ndarray

2.) I guess the """class ndarray_inMrcFile(N.memmap): pass""" construct is to simplistic .... Could someone suggest a *minimal* definition, so that my attributes would be preserved ?

A few comments about the wiki page: 1) The example does not show the printed info, such as "__new__ received %s", in the example session

2) I don't understand, why in the example the "info" attribute is set in "__new__()", even though the text above says: """However, we need to keep in mind that any attribute that we define in the __new__ method will be shared among all the instances. If we want instance-specific attributes, we still need some specific initialization. We cannot use the __init__ method, as it won't be called. That's where __array_finalize__ comes to play."""

3) the text about "The __array_finalize__ method" should at least once say that it is defined as "def __array_finalize__(self,obj)" -- otherwise I could only guess where the "obj" comes from.

4) related to the text after: "In other terms, __array_finalize__ is called" How do I know if a function or method actually invokes __new__ ;-) ? Would I have to study the numpy source ?

5) For the "__array_finalize__ method" there a text that says: """Subclasses inherit a default implementation of this method that does nothing""" --- but how about "__new__()" : what happens if you don't define that (either) ?

(Hope these comments are helpful)

Thanks for the help, Sebastian

First try seems to show that just changing my class def to: class ndarray_inMrcFile(N.memmap): def __array_finalize__(self,obj): self.Mrc = getattr(obj, 'Mrc', None) Seems to add the wanted attribute back into result of transpose(). However now I get (many!) exceptions like: Exception exceptions.AttributeError: "'ndarray_inMrcFile' object has no attribut e '_mmap'" in <bound method ndarray_inMrcFile.__del__ of ndarray_inMrcFile([ 6,.... Do I need to call super.__array_finalize__(obj) first ? -Sebastian

On Nov 23, 2007 5:01 PM, Sebastian Haase <haase@msg.ucsf.edu> wrote:

On Nov 23, 2007 4:43 PM, Sebastian Haase <haase@msg.ucsf.edu> wrote:

...

On Nov 23, 2007 3:37 PM, Pierre GM <pgmdevlist@gmail.com> wrote:

...

On Friday 23 November 2007 03:25:37 Sebastian Haase wrote:

...

Hi, this question might habe been answered before: I have my own ndarray-derived class. I did this, so that I can add another "custom attribute" -- let's say arr.filename

Sebastian, Could you post the __new__ and __array_finalize__ for your class ? Are you sure you update the attribute in __array_finalize__ ? And this page may come handy: http://www.scipy.org/Subclasses

Aah - I guess I did not do my homework ;-) Here is my code: class ndarray_inMrcFile(N.memmap): pass

data = self.data data.__class__ = ndarray_inMrcFile

Two points: 1.) Please don't get distracted by the way how I change the class type "after the fact". I get the data as a memmaped slice from a file. Then I do some (potentially not entirely clean) acrobatic to have this slice change it's class-type so that I can attach the original memmap (+ other attributes, such as filename) onto the ndarray

2.) I guess the """class ndarray_inMrcFile(N.memmap): pass""" construct is to simplistic .... Could someone suggest a *minimal* definition, so that my attributes would be preserved ?

A few comments about the wiki page: 1) The example does not show the printed info, such as "__new__ received %s", in the example session

2) I don't understand, why in the example the "info" attribute is set in "__new__()", even though the text above says: """However, we need to keep in mind that any attribute that we define in the __new__ method will be shared among all the instances. If we want instance-specific attributes, we still need some specific initialization. We cannot use the __init__ method, as it won't be called. That's where __array_finalize__ comes to play."""

3) the text about "The __array_finalize__ method" should at least once say that it is defined as "def __array_finalize__(self,obj)" -- otherwise I could only guess where the "obj" comes from.

4) related to the text after: "In other terms, __array_finalize__ is called" How do I know if a function or method actually invokes __new__ ;-) ? Would I have to study the numpy source ?

5) For the "__array_finalize__ method" there a text that says: """Subclasses inherit a default implementation of this method that does nothing""" --- but how about "__new__()" : what happens if you don't define that (either) ?

(Hope these comments are helpful)

Thanks for the help, Sebastian

First try seems to show that just changing my class def to: class ndarray_inMrcFile(N.memmap): def __array_finalize__(self,obj): self.Mrc = getattr(obj, 'Mrc', None)

Seems to add the wanted attribute back into result of transpose(). However now I get (many!) exceptions like: Exception exceptions.AttributeError: "'ndarray_inMrcFile' object has no attribut e '_mmap'" in <bound method ndarray_inMrcFile.__del__ of ndarray_inMrcFile([ 6,....

Do I need to call super.__array_finalize__(obj) first ?

-Sebastian

This seems to work without any problem now: class ndarray_inMrcFile(N.ndarray): def __array_finalize__(self,obj): self.Mrc = getattr(obj, 'Mrc', None) Comments? -Sebastian

On Nov 23, 2007 5:18 PM, Pierre GM <pgmdevlist@gmail.com> wrote:

...

This seems to work without any problem now: class ndarray_inMrcFile(N.ndarray): def __array_finalize__(self,obj): self.Mrc = getattr(obj, 'Mrc', None)

Comments?

That should work if you want a subclass of ndarray. That probably won't if you want a subclass of memmap. Once again, I'd do a def __new__(**options) N.memmap.__new__(**options) or something to this effect...

One more question on this: If I really don't need the memmap features explicitely anymore, and decide to derive from N.ndarray: Is it correct to assume that N.ndarray does *not* do it's own N.ndarray.__new__(**options) or so, that I would have to call ? (In other words: If I derive from N.ndarray I do not need to do a def __new__(**options): N.ndarray.__new__(**options) , right?) -Sebastian

2.) I guess the """class ndarray_inMrcFile(N.memmap): pass""" construct is to simplistic .... Could someone suggest a *minimal* definition, so that my attributes would be preserved ?

Mmh, I would at least try to explicitly call the N.memmap.__new__

A few comments about the wiki page: 1) The example does not show the printed info, such as "__new__ received %s", in the example session

OK, I'll edit the example

2) I don't understand, why in the example the "info" attribute is set in "__new__()", even though the text above says: """However, we need to keep in mind that any attribute that we define in the __new__ method will be shared among all the instances. If we want instance-specific attributes, we still need some specific initialization. We cannot use the __init__ method, as it won't be called. That's where __array_finalize__ comes to play."""

Yeah, that's not very clear. Well, look at the code of the example. In line 11, we call a view of the array, with our class as argument. That calls __array_finalize__, with "subarr" as the obj argument. __array_finalize__ sets a default "info" attribute: an empty dictionary, as the argumnet doesn't have an info attribute yet. In lines 14-18, we set the "info" argument of our object to the value we want.

3) the text about "The __array_finalize__ method" should at least once say that it is defined as "def __array_finalize__(self,obj)" -- otherwise I could only guess where the "obj" comes from.

OK, same as for 1.

4) related to the text after: "In other terms, __array_finalize__ is called" How do I know if a function or method actually invokes __new__ ;-) ? Would I have to study the numpy source ?

As a rule of thumb, __new__ is called each time you have a construction like MyClass(whatverargumentisneeded). When you rely on views (such as w/ the .view method, or ravel, or transpose...), you call __array_finalize__ . You don't really have to study the numpy code (even if it helps)

5) For the "__array_finalize__ method" there a text that says: """Subclasses inherit a default implementation of this method that does nothing""" --- but how about "__new__()" : what happens if you don't define that (either) ?

Try. The ndarray doesn't have a **keywords option: you can't pass extra parameters to ndarray.__new__. In other terms, if you need to pass some specific parameter to your class (in our example, info), you'll choke ndarray.