note about figures

ch3.tex

@@ -30,8 +30,44 @@ It is sometimes convenient (even recommended by some journals, including Cell) t
 \subsection{Performance on real codebases}
 \section{\sloppy NeuroticHelper improves neural network learning}
 
+\section{Graphics and figure quality}
+
+No matter how great the text content of your thesis is, the pictures will always catch the attention first. This creates the very important first impression of the thesis contents and general quality. Crucially, that also decides whether the thesis is later read with joy, or carefully examined with suspicion.
+
+Preparing your thesis in a way that this first impression gets communicated smoothly and precisely helps both the reviewer and you: the reviewer will not have a hard time understanding what exactly you wanted to convey, and you will get a better grade. Making the graphics `work for you' involves doing some extra work that is often unexpected. At the same time, you will need to fit into graphics quality constraints and guidelines that are rarely understood before you actually see a bad example.
+
+As a rule of thumb, you should allocate at least the same amount of time and effort for making the figures look good as you would for writing, editing and correcting the same page area of paragraph text.
+
+\subsection{Visualize all important ideas}
+The set of figures in your thesis should be comprehensive and complete. For all important ideas, constructions, complicated setups and results there should be a visualization that the reader can refer to in case the text does not paint the `mental image' sufficiently well. At the bare minimum, you should have at least 3 figures (roughly corresponding to the 3 chapters) that clearly and unambiguously show:
+\begin{enumerate}
+\item the context of the problem you are solving, optionally with e.g.~question marks and exclamation marks placed to highlight the problems and research questions
+\item the overall architecture of your solution (usually as a diagram with arrows, ideally with tiny toy examples of the inputs and outputs of each box),
+\item the advancement or the distinctive property of your solution, usually in a benchmark plot, or as a clear demonstration and comparison of the results.
+\end{enumerate}
+
+\subsection{Make the figures comprehensible}
+The figures should be easily comprehensible. Surprisingly, that requires you to follow some common ``standards'' in figure design and processing. People are often used to a certain form of the visualizations, and (unless you have a very good reason) deviating from the standard is going to make the comprehension much more complicated. The common standards include the following:
+\begin{itemize}
+  \item caption everything correctly, place the caption at an expectable position
+  \item systematically label the plots with `main' titles (usually in boldface, above the plot), plot axes, axis units and ticks, and legends
+  \item lay out the diagrams systematically, ideally follow a structure of a bottom-up tree, a left-to-right pipeline, a top-down layered architecture, or a center-to-borders mindmap
+  \item {use colors that convey the required information correctly \par\footnotesize Although many people carry some intuition for color use, achieving a really correct utilization of colors is often very hard without previous experience in color science and typesetting. Always remember that everyone perceives color hues differently, therefore the best distinction between the colors is done by varying lightness of the graphics elements (i.e., separating the data by dark vs.~light) rather than by using hues (i.e., forcing people to guess which one of salmon and olive colors means ``better''). Almost 10\% of the population have their vision impaired by some form of color vision deficiency, most frequently by deuteranomaly that prevents interpreting even the most `obvious' hue differences, such as green vs.~red. Finally, printed colors look surprisingly different from the on-screen colors. You can prevent much of these problems by using standardized palettes and well-tested color gradients, such as the ones from ColorBrewer\footnote{\url{https://colorbrewer2.org}} and ViridisLite\footnote{\url{https://sjmgarnier.github.io/viridisLite/}}. Check if your pictures still look good if converted to greyscale, and use a color deficiency simulator to check how the colors are perceived with deuteranomaly.}
+\end{itemize}
+
+Avoid large areas of over-saturated and dark colors:
+\begin{itemize}
+  \item under no circumstances use dark backgrounds for any graphical elements, such as diagram boxes and tables --- use very light, slightly desaturated colors instead
+  \item avoid using figures that contain lots of dark color (as a common example, heatmaps rendered with the `magma' color palette often look like huge black slabs that are visible even through the paper sheet, thus making a dark smudge on the neighboring page)
+  \item increase the brightness of any photos to match the average brightness of the text around the figure
+\end{itemize}
+
+Remember to test your figures on other people --- usually, just asking `What do you think the figure should show?' can help you debug many mistakes in your graphics. If they think that the figure says something different than what you planned, then most likely the figure is wrong, not the understanding of others.
+
+Finally, there are many magnificent resources that help you arrange your graphics correctly. The two books by Tufte~\cite{tufte1990envisioning,tufte1983visual} are arguably classics in the area. Additionally, you may find many interesting resources to help you with technical aspects of plotting, such as the \texttt{ggplot}-style `Fundamentals' book by~\citet{wilke2019fundamentals}, and a wonderful manual for the TikZ/PGF graphics system by~\citet{tantau2015tikz} that will help you draw high-quality diagrams (like the one in~\cref{fig:schema}).
+
 \section{What is a discussion?}
-After you present the results and show that your contribution works, it is important to \emph{interpret} them, showing what they mean for the more general public.
+After you present the results and show that your contributions work, it is important to \emph{interpret} them, showing what they mean for the more general public.
 
 Separate discussion sections are common in life sciences where ambiguity is common and intuition is sometimes the only thing that the authors have; exact sciences and mathematicians do not use them as often. Despite of that, it is nice to precisely set your output into the existing environment, answering:
 \begin{itemize}

refs.bib

@@ -25,3 +25,34 @@
   year={1989},
   publisher={Státní pedagogické nakladatelství}
 }
+
+@book{tufte1990envisioning,
+  title={Envisioning information},
+  author={Tufte, Edward R and Goeler, Nora Hillman and Benson, Richard},
+  year={1990},
+  publisher={Graphics press Cheshire, CT}
+}
+
+@book{tufte1983visual,
+  title={Visual display of quantitative information},
+  author={Tufte, Edward R},
+  year={1983},
+  publisher={Graphics press Cheshire, CT}
+}
+
+@book{wilke2019fundamentals,
+  title={Fundamentals of Data Visualization},
+  author={Wilke, Claus O},
+  year={2019},
+  publisher={O'Reilly Media, Inc.},
+  url={https://clauswilke.com/dataviz/},
+  isbn={9781492031086}
+}
+
+@techreport{tantau2015tikz,
+  title={The TikZ and PGF Packages (Manual for version 3.1.8b)},
+  author={Tantau, Tilll},
+  year={2020},
+  institution={Institut f{\"u}r Theoretische Informatik Universit{\"a}t zu L{\"u}beck},
+  url={http://mirrors.ctan.org/graphics/pgf/base/doc/pgfmanual.pdf}
+}