Difference between revisions of "Channel Builder"
(→Comments about homework T) |
(→Comments about homework T) |
||
| (3 intermediate revisions by the same user not shown) | |||
| Line 9: | Line 9: | ||
Putting in 0.5 leads to a situation where the measurement does not provide information. In this case the state posterior PMF converges to the "equilibrium distribution for the Markov chain," which is the unique PMF which is unchanged by the prediction. In our case this PMF is heavily weighted to STUCK. Note that if you run a simulation most of the time the channel is STUCK (because it is easy to get into this state but hard to get out). In the absence of any information your best guess is STUCK. Hence the model's behavior when data = 0.5 is repeated. | Putting in 0.5 leads to a situation where the measurement does not provide information. In this case the state posterior PMF converges to the "equilibrium distribution for the Markov chain," which is the unique PMF which is unchanged by the prediction. In our case this PMF is heavily weighted to STUCK. Note that if you run a simulation most of the time the channel is STUCK (because it is easy to get into this state but hard to get out). In the absence of any information your best guess is STUCK. Hence the model's behavior when data = 0.5 is repeated. | ||
| − | A datum of 0.5 is very unlikely for INoiselevel = 0.01 and leads to huge negative log-likelihoods. It is not so unlikely with a higher | + | A datum of 0.5 is very unlikely for INoiselevel = 0.01 and leads to huge negative log-likelihoods. It is not so unlikely with a higher noise level = 1. |
[[Media:Lab_T.pdf|A bonus problem for Lab T.]] Can you spot the problem in the answer for this Bonus Problem? | [[Media:Lab_T.pdf|A bonus problem for Lab T.]] Can you spot the problem in the answer for this Bonus Problem? | ||
| Line 15: | Line 15: | ||
== Today's Lab == | == Today's Lab == | ||
| − | + | Today's lab is the first Channel Builder Tutorial (Creating a channel from an HH-style specification) linked from [http://www.neuron.yale.edu/neuron/docs/chanlbild/main.html here]. | |
| + | |||
| + | '''Homework:''' Do the experiments in the tutorial, for each one copy and paste the response (voltage versus time trace) into a Powerpoint and send to me. | ||
| + | |||
| + | Warning: I found the directions tricky, so be careful. | ||
Latest revision as of 16:18, 14 April 2009
Comments about homework T
Putting in a '1' makes you certain that you are open.
Putting in a '1' then a '0' makes you certain that you are closed.
Putting in more '0's leads to greater and greater certainty that you are stuck.
Putting in 0.5 leads to a situation where the measurement does not provide information. In this case the state posterior PMF converges to the "equilibrium distribution for the Markov chain," which is the unique PMF which is unchanged by the prediction. In our case this PMF is heavily weighted to STUCK. Note that if you run a simulation most of the time the channel is STUCK (because it is easy to get into this state but hard to get out). In the absence of any information your best guess is STUCK. Hence the model's behavior when data = 0.5 is repeated.
A datum of 0.5 is very unlikely for INoiselevel = 0.01 and leads to huge negative log-likelihoods. It is not so unlikely with a higher noise level = 1.
A bonus problem for Lab T. Can you spot the problem in the answer for this Bonus Problem?
Today's Lab
Today's lab is the first Channel Builder Tutorial (Creating a channel from an HH-style specification) linked from here.
Homework: Do the experiments in the tutorial, for each one copy and paste the response (voltage versus time trace) into a Powerpoint and send to me.
Warning: I found the directions tricky, so be careful.
