Add section: visualising neural networks with seaborn and TensorBoard

[Bardzioch]

Introduces a new subsection in "Tables and figures" covering:

• architecture visualisation with torchviz and netgraph
• training monitoring with TensorBoard
• feature map extraction via forward hooks
• confusion matrix produced with seaborn (executable figure)

Adds seaborn to pip install cell and extends .wordlist.txt
with the required technical terms.

closes #73

[Bardzioch]

I have no idea why 1 check failed, some issue with rsvg-convert that didn't exist before, maybe someone encountered anything similar because i couldn't figure it out

[Bardzioch]

Video showcase: https://drive.google.com/file/d/1FmClkuEQg9e-RskZRiTxcOroYN-kiXr9/view?usp=sharing

[slayoo]

we've just watched the video, thanks

[slayoo]

reviewers: @Man1ek27, @Kamo3131

[Kamo3131]

• 👌 check if the contents match the topic and assigned issue comments (or a subset of them)
• 📸 paste into the PR discussion a screenshot of the reviewed section from the pdf generated on CI (action artifacts)
• 📄 check if the page count of the introduced section is within the 1/2 ... 2 page limit (optimum: 1 page)
• 🧠 check if the text is understandable for our target audience (incl. those who might be using Quarto, LaTeX, markdown, GitHub, etc for the first time)
• 💻 if the PR includes source code listings, check if these are understandable, complete and self-sufficient
  There are no codes listings.
• 📊 if there are any plots, check if the font sizes of all tick labels, legend keys, and axis titles are typeset with a font roughly matching the font size in the main body of the text
• 🖼️ if there are graphics or tables, check if these are numbered and annotated with captions
• ✂️ if there are any plot/graphics, check if these are vector graphics, and if the page loading time is not too long, or if the pdf size is not overly big
• 📂 if there are any auxiliary files added to the repo, check if their names and location matches our suggestions for folder structure (#112)
• ✏️ check if there are no typographic errors (#111)
• (N/A) if there are any equations, check if these are mathing the math typesetting guidelines (#87)
• 📐 check if there is no excessive white space introduced and if the graphics/tables/diagrams/listings fit within page margins
• 📑 check if the section title fits well into the TOC, and if the section location within the document makes sense
• 📖 check if the author added her/his name to the colophon
  Not added.
• 🏗️ check if the PR does not violate the K.I.S.S. principle - any unneeded files, lines, images, code lines, sentences, jargon, etc
• 🔀 check if git usage is sane (e.g., PRs should never originate from a main branch as it blocks further development in the fork)
• 🪟 check if code or text does not assume any particular operating system, and if it does it alternative solutions are explained

I think you should add some code in a code cell format, rather than just text, to the topic, because after reading the PDF, the reader won't know how to implement what you've written. It doesn't have to be a whole block of code, just the essential functions.

[Bardzioch]

I think you should add some code in a code cell format, rather than just text, to the topic, because after reading the PDF, the reader won't know how to implement what you've written. It doesn't have to be a whole block of code, just the essential functions.

I'll try to point readers to this, because the 1.5 page size limit is very tough to satisfy.

[Man1ek27]

Good scope coverage overall: the section matches issue #73 and includes architecture visualisation, training monitoring context, feature maps, and confusion-matrix diagnostics.

Main required changes:

1. Add scipy to dependencies (the section uses from scipy.ndimage import convolve, but install cell currently adds only matplotlib seaborn netgraph).
2. Add one minimal executable snippet for TensorBoard / SummaryWriter (or a basic forward hook), so the section is actionable for beginners.
3. Add the section author to colophon/author metadata.

Minor:

• Keep technology-name formatting consistent (e.g., PyTorch vs \techword{...}).

[Bardzioch]

done i think i guess

[Man1ek27]

[slayoo]

one of the graphics is certainly not vector: