Datasets

I need a mathematically simplistic, multidimensional dataset that can be generated with machine accuracy.

I’ve been using [x,y] with truth determined by some hidden function, but it’s hard to generalize that beyond 3 dimensions. It does have the advantage of being accessible to machine limits, though.

Deep Learning

More dakka?

But what if…more dakka isn’t enough dakka?

Then…even more dakka!

On a related note, I’m running models with hundreds of thousands to millions of neurons a layer.

Principal Component Analysis

Sometimes I just start doing a little PCA and lose all sense of time.

It happens more than you’d think, and while that’s probably a low bar to get over, it’s kinda weird that it crops up as often as it does.

Stable Diffusion

Stable Diffusion does landscapes pretty well.

I’d like to use it to creat human models I can practice drawing, but the details are what SD gets wrong and what I need to get right.

I haven’t gotten any good shots of Castle Amber, FYI.

Feature Engineering

You want your features to induce four things: information, linear independence, resilience, and speed.

Example: you want to figure out who someone is.

Obviously, your features should have enough information that the model can accurately answer a question. A good feature is a name.

You don’t want redundant features. The first character of the name is highly linearly dependent with the name. Not perfectly, as you could get noise on the line, a misclick, or something similar, but very, very dependent. The first character of the name would be less useful if the name is another feature.

You don’t want a system so light that noise in any one feature throws the model off. So you don’t only want the name as a feature because people change their names, use nicknames, and so forth. If all you’ve got is the name, and Thomas is going by Tom, you’ve got a problem. A problem that the first letter of the name as a name might solve.

Computers don’t know the connection between Thomas and Tom, much less Elizabeth and Liz. So you’ve got to make judgement calls. But a balance of the three aspects such that the programmer maximizes the norm is a good direction to start.

Of course, you also want your model to work on finite hardware, so speed maximization is a factor. Maximize norm(I, LI, R)*speed.