Programming Forums - View Single Post

skuinders · Jun 27th, 2005, 1:24 PM

See if you can solve this

http://crux.baker.edu/cdavis09/roses.html

DON'T POST THE ANSWER
warning: this may piss you off after a while of not getting it!

I got it after about 9 minutes... I know some people in other forums who have spent days on it. After I solved it, I tried to see if a neural network could find a way to generate correct petal counts.

I encoded the dice roll data as follows:
N = number of rolls
ND = number of dice in each roll

the input is a NxND matrix of random values from 1 to 6
the class (desired output) vector holds the petal count corresponding to each row (roll) of the input matrix

using a simple backpropagation NN, it is not possible to solve this problem as encoded the above way because of the inherent non-linearity of the problem. (if you plot the 2D case, you can clearly see that the petal counts are not linearly separable and a simple non-linear function is not immediately clear either).

there are other (more trivial) ways you could encode the roll information and make it work, but such models would be of little use.

I know I beat things to death.... anyways, have fun!

Jun 27th, 2005, 1:24 PM	#1
skuinders Hobbyist Programmer Join Date: Jun 2005 Location: MA, US Posts: 204 Rep Power: 4	Petals Around the Rose See if you can solve this http://crux.baker.edu/cdavis09/roses.html DON'T POST THE ANSWER warning: this may piss you off after a while of not getting it! I got it after about 9 minutes... I know some people in other forums who have spent days on it. After I solved it, I tried to see if a neural network could find a way to generate correct petal counts. I encoded the dice roll data as follows: N = number of rolls ND = number of dice in each roll the input is a NxND matrix of random values from 1 to 6 the class (desired output) vector holds the petal count corresponding to each row (roll) of the input matrix using a simple backpropagation NN, it is not possible to solve this problem as encoded the above way because of the inherent non-linearity of the problem. (if you plot the 2D case, you can clearly see that the petal counts are not linearly separable and a simple non-linear function is not immediately clear either). there are other (more trivial) ways you could encode the roll information and make it work, but such models would be of little use. I know I beat things to death.... anyways, have fun! __________________ "A stupid man's report of what a clever man says can never be accurate, because he unconciously translates what he hears into something he can understand." - B. Russell http://web.bryant.edu/~srk2 Last edited by skuinders; Jun 27th, 2005 at 1:32 PM.