1
00:00:03,000 --> 00:00:11,000
For the first question, the first step in the UC Santa Cruz prediction strategy, after they gathered the data, was to perform delay-coordinate embedding of the data
2
00:00:11,000 --> 00:00:15,000
It should be noted that they also used several other prediction strategies
3
00:00:15,000 --> 00:00:20,000
But, the one were referring to here: use delay-coordinate embedding as the first preprocessing step of the data
4
00:00:20,000 --> 00:00:28,000
For Question 2, a neural net-based method worked better than a nonlinear dynamics-based method on SFI dataset A in the long term
5
00:00:28,000 --> 00:00:29,000
And this is false
6
00:00:29,000 --> 00:00:37,000
While Wans method, which was the neural net method, worked great in the short term, Tim Sauers nonlinear dynamics-based method did far better in the long term
7
00:00:37,000 --> 00:00:38,000
So this question is false
8
00:00:38,000 --> 00:00:42,000
The third question is simply, How does Lorenzs method of analogues work?
9
00:00:42,000 --> 00:00:45,000
And this is near-neighbor prediction on the trajectory of a dynamical system
10
00:00:45,000 --> 00:00:56,000
So given a known trajectory of a dynamical system, and a point in that state space, it looks for the points nearest neighbor and the known trajectory, then uses the next point in the trajectory as the forecast of where that point will go next
11
00:00:56,000 --> 00:00:58,000
So this first answer is correct
12
00:00:58,000 --> 00:01:03,000
The third answer is a simple modification of Lorenzs method of analogues using k nearest neighbors
13
00:01:03,000 --> 00:01:07,000
However, the original Lorenzs method of analogues used one neighbor
14
00:01:07,000 --> 00:01:09,000
So the first answer is correct