|
| "Another curious circumstance is that, although induction has proved to be a tricky thing to understand and justify logically, the human mind has a predilection for rampant, uncontrolled induction, and it requires much education to overcome this." -- E.T. Jaynes
| | |
| Things I'd like to learn about right now1) Variational methods & particle filters.
2) Conjugate priors.
3) Subtleties of MCMC.
If you'd like to talk about any of the above, lemme know.
| | |
| Isaiah Berlin on causality and free will"'Power' and 'accident' are but names for ignorance of the causal chains, but the chains exist whether we feel them or not; fortunately, we do not; for if we felt their weight, we could scarcely act at all; the loss of the illusion would paralyse the life which is lived on the basis of our happy ignorance. But all is well: for we never shall discover all the causal chains that operate: the number of such causes is infinitely great, the causes themselves infinitely small..." (Foxes and Hedgehogs)
| | |
|
I tend not to trust product reviews very much. I don't trust expert
reviews because
* I don't know if the expert prioritizes what I do; (s)he might be picky about certain things etc.
* I don't know whether or not the expert is glossing over or ignoring certain faults because it's bad business to do so.
* I often don't know whether the review is actually an expert, in which case the things I mention below apply.
As for Joe-Schmoe reviews (some of these apply but less-so IMO to experts):
* Non-pros who review stuff have bought it, and they're apt to seek
to justify their decisions. The more Jane spends as a proportion of
her income, the more likely it is that she'll inflate her view of the
prodict, I imagine.
* People who review stuff tend not to have musch in the way of
comparison data points -- they're apt to be comparing a product to
whatever they had before or after. This could be a source of
systematic error.
* People may modify their opinions to reflect their opinions of brands.
Anyway, it seems like a "blind use test" ould be a great way to
examine interactions of well-known phenomena in social psych in the
real world as well as being really helpful for consumers--one could
control for the sources of most of the above variation. Has anyone
done this sort of thing? I imagine it'd be somewhat pricy to start;
20k+ for equipment, but the marginal cost per-participant might be
pretty small, and I bet there are industry types who'd want to fund
such a thing, e.g., execs who thing their products have an undeservedly poor rep.
| | |
| TestHere's a little test. Basically, there are a bunch of strings of 6 characters below, and there's a pattern behind the identities of the characters. The identities of the characters on a given line are independent of the characters that appear of any other line; each line is sampled from an underlying distribution. The goal is to use the data to fill in the blanks on the incomplete strings below. None of the strings in the test block are identical to any of the strings in the training block. I'll buy you a drink if you solve it and tell me how did it. Extra points if you can tell me how the strings were generated. I think it can be done without using any tools other than your brain, and I think there's more training data than you strictly need.
J W Z V T A
J F Z V G E
U W D M L A
Y P S I O B
Y W S I T C
J P Z V L B
J P Z V T C
J F Z V T B
J W Z V G C
Y F S I L C
J P Z V T C
Y W S I T C
U P D M G B
J W Z V G C
J P Z V O E
U W D M T B
J R Z V T E
Y F S I O A
H R K N L B
U F D M L B
J P Z V G A
H F K N O B
U W D M G E
U F D M O E
Y P S I G C
J F Z V O C
U F D M L B
Y F S I L C
U P D M O A
Y F S I G B
U F D M O E
Y P S I O B
J P Z V L B
Y R S I G E
Y P S I G C
H W K N T E
U P D M L C
J W Z V L E
U F D M G A
H W K N O A
Y P S I G C
J R Z V T E
U R D M G C
U F D M L B
U R D M T A
H W K N L C
U P D M O A
H R K N O E
U R D M T A
J F Z V L A
U P D M L C
H W K N T E
Y F S I O A
J R Z V O A
H W K N O A
U F D M T C
Y P S I T A
Y F S I G B
J F Z V O C
H P K N T B
H R K N G A
H R K N O E
Y W S I O E
H W K N T E
H W K N O A
J R Z V G B
U F D M O E
U F D M T C
Y F S I T E
U W D M G E
U P D M O A
U P D M L C
H W K N T E
Y W S I G A
U R D M T A
U P D M G B
H R K N O E
Y F S I L C
H W K N O A
U F D M G A
U P D M G B
H P K N G E
H P K N G E
Y P S I O B
Y R S I O C
H P K N L A
Y W S I T C
H F K N G C
U W D M O C
Y F S I O A
--------Generalization test--------
Y R S _ T B
_ _ _ _ T E
H _ _ N O _
_ R S _ _ _
Y R S I L _
_ P _ I _ _
_ _ _ M T _
_ _ S I L _
H W K N G _
_ P _ _ L _
| | |
|
