Getting the docs up and running #225
Conversation
| matrix. In Gradient Descent, $P$ is simply an appropriately dimensioned identity matrix. | ||
| This means that we go in the exact opposite direction of the gradient. This means | ||
| that we do not use the curvature information from the Hessian, or an approximation | ||
| of it. While it does seem quite logical to go in the opposite direction of the fastest |
There was a problem hiding this comment.
Repetition of "This means" in two consecutive sentences.
Current coverage is 84.17%@@ master #225 diff @@
==========================================
Files 30 30
Lines 1895 1896 +1
Methods 0 0
Messages 0 0
Branches 0 0
==========================================
Hits 1596 1596
- Misses 299 300 +1
Partials 0 0
|
|
|
||
| $ m_k(s) = f(x_n) + \nabla f(x_n)^\top \textbf{s} + \frac{1}{2} \textbf{s}^\top H(x_n) \textbf{s} $ | ||
|
|
||
| For functions where $H(x_n)$ is difficult, or computationally expensive, to optain, we might |
There was a problem hiding this comment.
-> "or computationally expensive to obtain"
|
I think Documenter might have been updated :) It looks different now, but in a good way. |
|
Commented out some sections I hadn't written yet. I need to be a bit more sure what is going on in conjugate gradient to formulate something sensible. I know how it works, but I'm not sure how to present it in a paragraph or two. |
|
I'm going to push a version of this asap, and then the missing solvers can be added over the summer. I am just trying to find out how to get the travis-things for Documenter set without admin rights :) |
|
Let's see if it works (and if it doesn't, hope that Hatherly is not too busy with JCon :)) |
It's easier to comment on the docs inline if there's a PR, as mentioned in #224 .