search

ViktorWase / Frank-the-Science-Bot

Frank is a bot that does science. He takes a dataset and forms hypothesises about mathematical constants (aka conservation laws) of the data. He then tests the hypothesis.
Apache License 2.0
2 stars 0 forks source link

howMuchOfATautologyItIs #6

Open ViktorWase opened 9 years ago

ViktorWase commented 9 years ago

This function should take weight vector (which represents a function in combination with an ANN) and see if it always spits out zero. Because the ANN-function should return close to zero for all data points in the database. But not just for random crap.

The function 0 x mass returns zero for all values of mass, while (mass x velocity^2)/2-constant doesn't. If the first one is true for the database it gives no scientific insight into anything. If the other is true and returns zero for all data points it means that kinetic energy is conserved in the system. Which might actually be useful.

So this function should return a high value (one?) if it's "close" to being a useless tautology. And a low value (zero?) if it's not.

johnbrynte commented 9 years ago

What should be considered a tautology? Is it only to remove functions that are constant when fed with random crap? If so then you only need to write a function that runs the function with a number of random values, which in turn decides the probability of a tautology. If that function does not always give the same value then it's not a tautology, otherwise it might be with a certain probability.

ViktorWase commented 9 years ago

Yeah, that seems like a clever idea. Do it!

ViktorWase commented 9 years ago

But it might not always be exactly the same number. Because 0.0000001 x m = 0 is "closer" to being a tautology than, say, 2 x m = 0.

ViktorWase commented 9 years ago

I did a temporary version of this function. But it needs fixing.