Equivalency between 2 models/agents.

[MridulS]

Should HARK have the __eq__ method to compare 2 HARK objects?

For example

example_1 = PerfForesightConsumerType(**PF_params)
example_2 = PerfForesightConsumerType(**PF_params)

example_1 == example_2 should be True (it will give it False) but currently HARK doesn't have the ability to test this.

[sbenthall]

That sounds nice.

A couple thoughts:

• The distanceMetric functionality seems to do some of this. That functionality is old code. It could probably be reimplemented in a more pythonic way. Maybe that would be connected to this issue.
• Would an unsolved and a solved AgentType be equivalent? What if a simulation had been ran, with different parameters? The meaning of 'equivalence' might be more intuitive if some of the functionalities currently handled in AgentType were disentangled, see Disentangle solution and simulation frameworks in HARK.core and downstream classes #495

[MridulS]

Would an unsolved and a solved AgentType be equivalent? What if a simulation had been ran, with different parameters? The meaning of 'equivalence' might be more intuitive if some of the functionalities currently handled in AgentType were disentangled, see #495

Different parameters should mean the agent are not equivalent. There isn't any way to get all the parameters of an agent right? something like AgentType.parameters

[mnwhite]

distanceMetric is used for comparing solutions, to see if it has converged on successive iterations. I'm not sure there's a use case for comparing instances of AgentType subclasses for equality.

…

On Tue, Apr 7, 2020 at 10:23 AM Mridul Seth ***@***.***> wrote: Would an unsolved and a solved AgentType be equivalent? What if a simulation had been ran, with different parameters? The meaning of 'equivalence' might be more intuitive if some of the functionalities currently handled in AgentType were disentangled, see #495 <#495> Different parameters should mean the agent are not equivalent. There isn't any way to get all the parameters of an agent right? something like AgentType.parameters — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#612 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADKRAFPJFZRL6XBUX3JFSRDRLMZM7ANCNFSM4MCXMCXQ> .

[sbenthall]

There isn't any way to get all the parameters of an agent right? something like AgentType.parameters

I don't believe there is.
But this is high on my refactoring wishlist.

It's a little complicated to do this... see #492

[sbenthall]

#760 #761 are related

[sbenthall]

See also #660

[sbenthall]

distanceMetric is used for comparing solutions, to see if it has converged on successive iterations. I'm not sure there's a use case for comparing instances of AgentType subclasses for equality.

This makes sense to me.

I have wondered for a while why AgentType inherits from HARKObject, since there is no use case for using its distanceMetric function.

I think what it's using currently is assignParameters.

Does assignParameters ever get used on a solution object? If not, then I think it would be safe to split HARKObject into two classes.

It would be safe to do that anyway, since the two classes can be combined with multiple inheritance when necessary.