christmas present

[From Bill Powers (941227.1500 MST)]

Bruce Abbott (941227.1250 EST)--

Good news about the animal experiments!

An 8088 is a real limitation, but I did lots of experiments on one with
only 640K total memory.

For the initial experiment replicating the Motheral data I think we can
get by pretty simply. While we want to know the time of each event, I
think we can avoid having to store times for each data point. An easy
way would be to write one byte to a 64-K array on the heap every 0.2
second. Each byte would consist of two four-bit units representing data
at 0.1-sec intervals, each unit consisting of

Bit 0: lever down
Bit 1: reinforcement triggered
Bit 2: position 1 (animal at bar)
Bit 3: position 2 (animal at food dish)

This would give us 128000 four-bit units recorded in 12800 seconds or 3
hr. 20 min. Every 3 hr 20 min we would block-write the (time-stamped)
array to the disk and clear the array for the next segment. A brief
pause for writing to disk every three hours won't affect the model,
especially if we record the starting time at the beginning of each
array. A high-density diskette will hold 24 hours'-worth of data. Did I
calculate this right?

I think that 0.1-sec resolution should be plenty. You might think of
covering the case of bar-presses at more than 10 per second by reporting
only one press in each tenth of a second, storing added presses to be
reported in successive tenths. I trust that bursts this fast will be
quite rare, so the slight distortion of the times will not have any
important effects.

A series of randomized FR experiments would make a nice package,
covering the range of 1:1 to about 160:1 that Motheral covered. I
suppose that to make it a real replication we should use all the
standard conditions: 80% body weight, experiments done in daily
sessions, feeding adjusted outside experimental conditions to maintain
weight, and so forth. I don't know about using "access to reinforcement"
rather than just delivering pellets -- seems to me that the "access"
approach leaves the "rate of reinforcement" actually received pretty
fuzzy. But that's what Motheral did. I trust you'll be measuring
original quantities and spillage -- once or twice a day should be enough
to give reasonable estimates of actual consumption, pro-rated per
reinforcement.

Let's also be thinking about how to get a faster computer -- even a
486SX-25 would be something like 40 times as fast. Maybe we can take up
a collection.

Funny how easy it is to say "reinforcement" when I don't even believe in
it.

ยทยทยท

--------------------------------------------------------------------
Best,

Bill P.