Topographica has GlobalParams (should be called ScriptParams) that allows parameters to be added to a script. Would need cleaning up.

Have patch from Marco (mixed with logging changes).

Converted from SourceForge issue 2888861, submitted by ceball
Submit Date: 2009-10-29 15:43 GMT

set e.g. .overrides on Parameterized instance, and override getattribute to look in .overrides first; would have to set .overrides=None after call. Or something like that.

Just need to run distutils on windows machine

Converted from SourceForge issue 3440912, submitted by ceball
Submit Date: 2011-11-22 00:25 GMT

Maybe it's obscure, but if in a script I have a global param "someparam" (p.someparam below) and I also make a variable "someparam", I get a warning.

[jupiter1]s0454615: cat anothertest.ty
print "Executing test.ty"

from topo.misc.commandline import global_params as p
import param

p.add(
someparam = param.String(default='1')
)

print p.someparam

someparam = "hello"

[jupiter1]s0454615: ./topographica anothertest.ty
Executing test.ty
1
Time: 000000.00 global_params: Warning: 'someparam=hello' is unused.

As a suggestion, it might be worth offering a different way of accessing a parameters value generator than the current get_value_generator interface.

Instead of using:

imagen.Gaussian(orientation=UniformRandom()).get_value_generator('orientation') 

It would be useful to get the generator using:

imagen.Gaussian(orientation=UniformRandom()).params('orientation').value_generator 

As the params method is probably the most common interface for accessing parameters and their attributes. This suggestion is related to issue #66.

Installing param (pip install param) into a virtual environment that is inside a git repository fails with

Exception: Cannot find any git version tags of format v_._

This makes it difficult to install param as part of certain Jenkins CI workflows that make use of git + virtualenv.

Converted from SourceForge issue 2896801, submitted by ceball
Submit Date: 2009-11-12 19:33 GMT

ceball@fiver:/working/topographica$ cat test.ty
name=True
ceball@fiver:/working/topographica$ ./topographica test.ty
Time: 000000.00 global_params: Warning: 'name=True' is unused.
ceball@fiver:~/working/topographica$

Converted from SourceForge issue 2333811, submitted by ceball
Submit Date: 2008-11-23 17:24 GMT

Converted from SourceForge issue 1829846, submitted by ceball
Submit Date: 2007-11-11 08:34 GMT

To allow declaration of something but not let it be set anywhere else,
even in a ParameterizedObject constructor.

Converted from SourceForge issue 2513082, submitted by ceball
Submit Date: 2009-01-16 12:02 GMT

E.g.

class Dynamic(Parameter)
def init(self,**params):
super(Dynamic,self).init(**params)

The name of **params is confusing here. It should be **kw - or in fact anything except **params, which is a name we use specifically for passing parameter values to methods of Parameterized instances.

Converted from SourceForge issue 2888857, submitted by ceball
Submit Date: 2009-10-29 15:41 GMT

err in defn in globparam.add() - error msg fails because attrname not set

Converted from SourceForge issue 3027874, submitted by ceball
Submit Date: 2010-07-10 18:48 GMT

r11145 added repr() for Parameter, so that the documentation of parametertized classes and instances could include type, bounds, default value and so on. Later reverted this change because it results in too much information in the gui (pop-up boxes), and probably too much even in the commandline help. Would be nice to have some compromise, though.

Converted from SourceForge issue 2630230, submitted by ceball
Submit Date: 2009-02-23 14:08 GMT

Rather than a text box, should be a combobox.

E.g. overridable_property https://github.com/ioam/param/blob/master/param/parameterized.py#L1673

Apparently there are too many ways.

Do we need a way that takes precedence as an argument, to allow filtering?

Converted from SourceForge issue 1374683, submitted by ceball
Submit Date: 2005-12-06 18:11 GMT

When I type help(my_object), I think it should at least
tell me that if there is no documentation here, typing
help(my_object_class) would probably provide what I'm
looking for.

Converted from SourceForge issue 1873416, submitted by ceball
Submit Date: 2008-01-17 04:01 GMT

We can override that in specific cases.

Converted from SourceForge issue 2545565, submitted by ceball
Submit Date: 2009-01-29 14:18 GMT

e.g.

[goshawk]v1cball: ./topographica -c name=4 -i
Time: 000000.00 global_params: Warning: 'name=4' is unused.

(name is a parameter of param.Parameterized)

Converted from SourceForge issue 1864370, submitted by ceball
Submit Date: 2008-01-05 07:58 GMT

Currently, we can't use python's iterators or generators with Dynamic parameters.

This is because:

(1) we deepcopy parameter values; python currently doesn't handle deepcopy of iterators and generators

(2) we want to be able to pickle simulations; python can't currently pickle iterators and generators (same as reason as 2)

(3) generators and iterators can't have attributes; we need to store _Dynamic_last and _Dynamic_time on the value generators

When creating a Parameterized instance, you can pass only **params:

class Parameterized(object):
    ...
    def __init__(self,**params):
        ...

ParameterizedFunction.instance() creates a Parameterized instance, so should have the same signature. Currently it looks like this:

class ParameterizedFunction(Parameterized):
...
    @bothmethod
    def instance(self_or_cls,*args,**params):
        ...

Also, passing *args to Parameterized.new (object.new) is at best pointless.

One thing that has frequently aggravates me about param is that there is no way (that I know of) to get at a dynamic value generator e.g a numbergen object. Perhaps there is a proper way to do this, but I have hacked in this method into my copy of param (on the Dynamic class) for the time being:

    def get_generator(self, obj):
        """
        Given a parameterized object, gets the dynamic value generator.
        """
        return super(Dynamic,self).__get__(obj, obj.__class__)

This is still awkward to use, but better than nothing. Say I have an imagen.Gaussian object and I want to fiddle the random number generator used for the orientation. Now I can do:

generator = gaussian.params('orientation').get_generator(gaussian)

Now for instance, I could change the time_fn for this numbergen object.

Any thoughts? Is there already a better way of doing this?

https://coveralls.io/r/ioam/param

Make sure it's accurately reporting coverage (see e.g. #14 (comment) and comment below it), then add a link/badge so we can easily see test coverage.

Implemented by Philipp in Topographica

From Jim:

We also [want] global parameters, with an object in the main namespace that
collects options specified on the commandline. I.e., any commandline option should
have a corresponding parameter, and changing that option should have a direct
mapping to that parameter, so that such parameters can be changed in a .rc file,
from within a script, etc.

From Jean-Luc:

Converted from SourceForge issue 3537800, submitted by jlstevens
Submit Date: 2012-06-25 14:49 GMT

[...] global parameters that are relevant to the whole simulator (eg. OpenMP
settings) would be very useful. [...] global_params should be implemented with the
following properties:

• At the commandline any option '-x=y' should set global parameter 'x' to value 'y'
• Global parameters belong to the global namespace so that you can put the line 'x=y'
  in your topographica.rc file to always set global parameter 'x' as 'y' by default. Any
  behaviour set in the rc file is overridden if explicitly set on the commandline.
• At the IPython prompt, entering 'x=y' should set global parameter 'x' to 'y' for the
  reminder of that session (ie. it should be possible to rebind global parameters at any
  time)

If this were implemented it would be possible to specify options in the topographica.rc
file instead of having to specify these options each time Topographica is invoke. All
the current options should be implemented this way (ie. -i,-v,-d,-l,-g, --pdb, -c, -p, -a)
with the exception of -h/--help.

(collected from #12 and #39)

Converted from SourceForge issue 3027504, submitted by ceball
Submit Date: 2010-07-09 18:10 GMT

We have our own warning system. How are e.g. tests supposed to detect our warnings?

If necessary (default may be ok).

See http://docs.travis-ci.com/user/notifications/

Dynamic parameters currently accept callable objects which are called to generate new values (e.g. with numbergen). It would make sense (and probably be more Pythonic!) to use the iterator protocol instead, calling the next function (for Python 3 compatibility) to get each value in turn.

It should be relatively easy to update numbergen to support the iterator protocol - it may be possible to deprecate the callable interface without too much trouble.

Also compare with Traits, Traitlets, etc

Hopefully make better use of doctest (inline)

Converted from SourceForge issue 1829837, submitted by ceball
Submit Date: 2007-11-11 08:16 GMT

Given how private attributes work in Python, it seems like we can just
have a parameter __abstract_class=True for abstract classes, and then
__is_abstract can check to see if there is such an attribute (and that
it is True, just in case).

I think https://github.com/ioam/param/blob/master/setup.py#L44 should just be if 'upload' in sys.argv.

Converted from SourceForge issue 1869380, submitted by ceball
Submit Date: 2008-01-11 15:12 GMT

Instead use (e.g. negative) precedence.

1. Update demo code.
2. Explain that code is much more complex than necessary, and document plans to wrap it in a simpler interface (with the aim of eventually replacing the original code with something much simpler).

We've been using printf-style formatting (https://docs.python.org/3/library/stdtypes.html#printf-style-string-formatting), but python now offers nicer ways to format strings.

Note that we can't use the following, because it's not available in 2:

Changed in version 3.1: The positional argument specifiers can be omitted, so '{} {}' is equivalent to '{0} {1}'

Converted from SourceForge issue 2888867, submitted by ceball
Submit Date: 2009-10-29 15:52 GMT

Slow opening of cfs, etc. deepcopy slot.

(https://mail.google.com/mail/?shva=1#search/%22Topographica-cvs+Digest%2C+Vol+30%2C+Issue+19%22/11dcade03c31c7ca)

Instead of our own.

Have patch from Marco (mixed with Python 3 changes).

Generalize and include in param or not?

Converted from SourceForge issue 3537800, submitted by jlstevens
Submit Date: 2012-06-25 14:49 GMT

Topographica has global_params which allow parameters to be specified with the -p flag. This naming is incorrect as global_params only behave properly within .ty scripts and global parameters that are relevant to the whole simulator (eg. OpenMP settings) would be very useful. The existing global_params should therefore be called script_params and global_params should be implemented with the following properties:

- At the commandline any option '-x=y' should set global parameter 'x' to value 'y'

• Global parameters belong to the global namespace so that you can put the line 'x=y' in your topographica.rc file to always set global parameter 'x' as 'y' by default. Any behaviour set in the rc file is overridden if explicitly set on the commandline.
• At the IPython prompt, entering 'x=y' should set global parameter 'x' to 'y' for the reminder of that session (ie. it should be possible to rebind global parameters at any time)

If this were implemented it would be possible to specify options in the topographica.rc file instead of having to specify these options each time Topographica is invoke. All the current options should be implemented this way (ie. -i,-v,-d,-l,-g, --pdb, -c, -p, -a) with the exception of -h/--help.

Using info sent by JLS.

State what they are.

Is it possible to make a Parameter that has to be set? I.e. if not specified on instance creation, an exception will be raised?