Retrieval time differences?

[asmaloney]

Each of the three frameworks uses different magnitudes for their latency factor.

Here are the formulae from the ACT-R manual:

(if chunk is activated)

$RT = Fe^{−( f *Ai )}$

• RT: The time to retrieve chunk in seconds
• Ai: The activation of chunk i which is being retrieved
• F: The latency factor parameter
• f: The latency exponent parameter

(if failed to activate)

$RT = Fe^{−( f *τ )}$

• RT: The time until the failure is noted in seconds.
• τ: The retrieval threshold parameter
• F: The latency factor parameter
• f: The latency exponent parameter

The defaults are as follows:

Framework	latency factor (F)	latency exponent (f)	retrieval threshold (τ)
ccm	0.05	(can't set it? *)	0.0
pyactr	0.1	1.0	0.0
vanilla	1.0	1.0	0.0

* The code doing the calculations doesn't seem to allow for this, so I guess it's fixed at 1.0?

How do we get comparable results if the frameworks seem to be using different magnitudes? If we change the latency factor then some test models fail to complete properly.

I'm sure I'm missing some basic understanding here...

[asmaloney]

To add more detail, here are the two places ccm uses latency in calculations. They are found in the Memory class (dm.py):

  def fail(self,request_number):
     if self.threshold is None: 
         time=self.maximum_time
     else:
         time=self.latency*math.exp(-self.threshold)
         if time>self.maximum_time: time=self.maximum_time 
     yield time
     if request_number!=self._request_count: return
     
     self.error=True
     self._buffer.clear()
     self.busy=False

  def recall(self,chunk,matches,request_number):
     self.finst.add(chunk)
     time=self.latency*math.exp(-chunk.activation)
     if time>self.maximum_time: time=self.maximum_time
     yield time
     if request_number!=self._request_count: return
     self._buffer.set(chunk)
     for a in self.adaptors: a.recalled(chunk)
     self.busy=False

And here's how pyactr calculates it (where latency_factor is passed in as model_parameters["latency_factor"]):

def retrieval_latency(activation, latency_factor, latency_exponent):
    """
    Calculate retrieval latency.
    """
    return latency_factor*(math.exp(-activation*latency_exponent))