name: 20240206-dataviz class: title, middle ## IFT 3710/6759 ## Projets (avancés) en apprentissage automatique #### .gray224[6 février 2024 - Session 9] ### .gray224[Visualisation des données] .smaller[.footer[ Slides: [alexhernandezgarcia.github.io/teaching/mlprojects24/slides/{{ name }}](https://alexhernandezgarcia.github.io/teaching/mlprojects24/slides/{{ name }}) ]] .center[ <a href="http://www.umontreal.ca/"><img src="../../../assets/images/slides/logos/udem-white.png" alt="Mila" style="height: 6em"></a> ] Alex Hernández-García (he/il/él) .footer[[alexhernandezgarcia.github.io](https://alexhernandezgarcia.github.io/) | [alex.hernandez-garcia@mila.quebec](mailto:alex.hernandez-garcia@mila.quebec)]<br> .footer[[@alexhg@scholar.social](https://scholar.social/@alexhg) [](https://scholar.social/@alexhg) | [@alexhdezgcia](https://twitter.com/alexhdezgcia) [](https://twitter.com/alexhdezgcia)] ??? - The class is going to be a mix of lecture and demonstration --- ## Format of the class and objective This class will be a combination of lecture and demonstration. The .highlight1[goal] is that by the end of the class: * You have learnt the core concepts of data visualisation. * You know some ingredients that make a _good_ figure. * You know some ingredients that make a _bad_ figure. --- ## Why does data visualisation matter? .center[<img src="../../../assets/images/teaching/mlprojects/dataviz/diff_depth_table.png" alt="Data visualisation" style="width:90%"/>] -- .center[<img src="../../../assets/images/teaching/mlprojects/dataviz/diff_depth.png" alt="Data visualisation" style="width:90%"/>] --- count: false ## Why does data visualisation matter? .left-column[ .center[<img src="../../../assets/images/teaching/mlprojects/dataviz/phast_tabforcenet.png" alt="ForceNet" style="width:60%"/>] .center[<img src="../../../assets/images/teaching/mlprojects/dataviz/phast_tabdpp.png" alt="D++" style="width:60%"/>] .center[<img src="../../../assets/images/teaching/mlprojects/dataviz/phast_tabschnet.png" alt="SchNet" style="width:60%"/>] ] .smaller[ .references[ Duval et al. (2022). [PhAST: Physics-Aware, Scalable, and Task-specific GNNs for Accelerated Catalyst Design](https://arxiv.org/abs/2211.12020). arXiv 2211.12020. ] ] -- .right-column[ .center[<img src="../../../assets/images/teaching/mlprojects/dataviz/phast_ablation.png" alt="Ablation figure" style="width:100%"/>] ] --- count: false ## Why does data visualisation matter? .center[<img src="../../../assets/images/teaching/mlprojects/dataviz/diff_depth.png" alt="Data visualisation" style="width:90%"/>] * Figures in scientific publications and technical reports are often the .highlight1[main support of the key results]. * Data visualisation has the potential of enabling the .highlight1[understanding] of complex and large numerical relationships .highlight1[at a glance] * Figures can create a successful .highlight1[communication channel between you], the author of a complex data analysis, .highlight1[and your audience]: .highlight1[data visualisation is akin to a metaphor to convey a complex idea] --- ## General ideas #### Common, but incorrect assumptions -- * Readers read the abstract, introduction and methods before seeing the figures.  -- * If I can interpret the figure, my readers can interpret the figures too.  -- #### Generally good ideas and indications * Many readers examine the figures first. * Figures should stand on their own and be self-explanatory. * Figures should be designed for a broad audience, not for yourself. * A likely successful approach is: 1. Identify the key messages. 2. Draw visualizations about them . 3. Write the paper or report around them. .references[ Credit of these guidelines is owed to [Tracey Weissberger](https://twitter.com/T_Weissgerber), expert in data visualisation and scientific communication. ] --- ## Ten simple rules for better figures .center[<img src="../../../assets/images/teaching/mlprojects/dataviz/tensimplerules.png" alt="Ten simple rules for better figures" style="width:90%"/>] .references[ Rougier et al. (2014). [Ten simple rules for better figures](https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1003833). PLOS Computational Biology. ] --- ## Ten simple rules for better figures ### 1. Know your audience > "\[I\]t is important to identify, as early as possible in the design process, the audience and the message the visual is to convey". A figure should stand by itself when seen by the target audience. It should contain all the relevant information the audience _needs_ to know. Different audiences know different things. * Is the figure for yourself and collaborators? * Is it for a specialised audience? * Is it for the general public? .references[ Rougier et al. (2014). [Ten simple rules for better figures](https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1003833). PLOS Computational Biology. ] --- ## Ten simple rules for better figures ### 2. Identify your message > "Only after identifying the message will it be worth the time to develop your figure". .right-column[ <figure style="text-align: center"> <img src="../../../assets/images/teaching/mlprojects/dataviz/superior_colliculus.png" alt="Foveal magnification at the superior colliculus" style="width:90%"/> <figcaption style="text-align: center; font-size: small">The superior colliculus (SC) is a brainstem structure at the crossroads of multiple functional pathway. There is an extreme foveal magnification in the projection from the retina onto the SC. .cite[(Rougier et al., 2014)]</figcaption> </figure> ] .left-column[ * How can a figure simplify the main message, which is otherwise likely to be hard to express in words or in a table with numbers? * A good figure do not only _represent_ the data, but is at the service of the main message. ] .references[ Rougier et al. (2014). [Ten simple rules for better figures](https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1003833). PLOS Computational Biology. ] --- ## Ten simple rules for better figures ### 3. Adapt the figure to the support medium > "Each \[medium (a poster, a monitor, a projection screen, a PDF)\] represents different physical sizes for the figure, but more importantly, each of them also implies **different ways of viewing and interacting with the figure**". <figure style="text-align: center"> <img src="../../../assets/images/teaching/mlprojects/dataviz/trajectories.png" alt="Different figures for different media" style="width:55%"/> <figcaption style="text-align: center; font-size: small">Different figures for different media. The figure on the left has been designed for a journal article; the figure on the right for an oral presentation .cite[(Rougier et al., 2014)]</figcaption> </figure> .references[ Rougier et al. (2014). [Ten simple rules for better figures](https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1003833). PLOS Computational Biology. ] ??? * During a presentation, figures are display for a short time, and can be explained by the speaker * Readers of an article have more time read a caption and understand more details from a figure. --- count: false ## Ten simple rules for better figures ### 3. Adapt the figure to the support medium > "Each \[medium (a poster, a monitor, a projection screen, a PDF)\] represents different physical sizes for the figure, but more importantly, each of them also implies **different ways of viewing and interacting with the figure**". .center[<img src="../../../assets/images/teaching/mlprojects/dataviz/daug0.png" alt="Different figures for different media" style="width:70%"/>] --- count: false ## Ten simple rules for better figures ### 3. Adapt the figure to the support medium > "Each \[medium (a poster, a monitor, a projection screen, a PDF)\] represents different physical sizes for the figure, but more importantly, each of them also implies **different ways of viewing and interacting with the figure**". .center[<img src="../../../assets/images/teaching/mlprojects/dataviz/daug1.png" alt="Different figures for different media" style="width:70%"/>] --- count: false ## Ten simple rules for better figures ### 3. Adapt the figure to the support medium > "Each \[medium (a poster, a monitor, a projection screen, a PDF)\] represents different physical sizes for the figure, but more importantly, each of them also implies **different ways of viewing and interacting with the figure**". .center[<img src="../../../assets/images/teaching/mlprojects/dataviz/daug2.png" alt="Different figures for different media" style="width:70%"/>] --- count: false ## Ten simple rules for better figures ### 3. Adapt the figure to the support medium > "Each \[medium (a poster, a monitor, a projection screen, a PDF)\] represents different physical sizes for the figure, but more importantly, each of them also implies **different ways of viewing and interacting with the figure**". .center[<img src="../../../assets/images/teaching/mlprojects/dataviz/daug3.png" alt="Different figures for different media" style="width:70%"/>] --- count: false ## Ten simple rules for better figures ### 3. Adapt the figure to the support medium > "Each \[medium (a poster, a monitor, a projection screen, a PDF)\] represents different physical sizes for the figure, but more importantly, each of them also implies **different ways of viewing and interacting with the figure**". .center[<img src="../../../assets/images/teaching/mlprojects/dataviz/daug4.png" alt="Different figures for different media" style="width:70%"/>] --- count: false ## Ten simple rules for better figures ### 3. Adapt the figure to the support medium > "Each \[medium (a poster, a monitor, a projection screen, a PDF)\] represents different physical sizes for the figure, but more importantly, each of them also implies **different ways of viewing and interacting with the figure**". .center[<img src="../../../assets/images/teaching/mlprojects/dataviz/daug5.png" alt="Different figures for different media" style="width:70%"/>] --- count: false ## Ten simple rules for better figures ### 3. Adapt the figure to the support medium > "Each \[medium (a poster, a monitor, a projection screen, a PDF)\] represents different physical sizes for the figure, but more importantly, each of them also implies **different ways of viewing and interacting with the figure**". .center[<img src="../../../assets/images/teaching/mlprojects/dataviz/daug6.png" alt="Different figures for different media" style="width:70%"/>] --- count: false ## Ten simple rules for better figures ### 3. Adapt the figure to the support medium > "Each \[medium (a poster, a monitor, a projection screen, a PDF)\] represents different physical sizes for the figure, but more importantly, each of them also implies **different ways of viewing and interacting with the figure**". .center[<img src="../../../assets/images/teaching/mlprojects/dataviz/daug7.png" alt="Different figures for different media" style="width:70%"/>] --- ## Ten simple rules for better figures ### 4. Captions are not optional > "The caption explains how to read the figure and provides additional precision for what cannot be graphically represented". * Figures can hardly explain everything by themselves. * Captions support the images on being self-explanatory. * Captions can further highlight the key message of the figure. .references[ Rougier et al. (2014). [Ten simple rules for better figures](https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1003833). PLOS Computational Biology. ] --- ## Ten simple rules for better figures ### 5. Do not trust the defaults > "\[Default settings of software or libraries\] are good enough for any plot but they are best for none". <figure style="text-align: center"> <img src="../../../assets/images/teaching/mlprojects/dataviz/defaults.png" alt="Default settings for plotting sinusoids" style="width:70%"/> <figcaption style="text-align: center; font-size: small">The default settings are clearly suboptimal for the plot on the left. Tuning the tick labels, the legend and labelling can greatly improve the figure.</figcaption> </figure> * Defaults are fine enough for _quick and dirty_ visualisation, but the final figures require careful design and optimisation. .references[ Rougier et al. (2014). [Ten simple rules for better figures](https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1003833). PLOS Computational Biology. ] --- ## Ten simple rules for better figures ### 6. Use colour effectively > "\[C\]olor can be either your greatest ally or your worst enemy if not used properly". .left-column[ <figure style="text-align: center"> <img src="../../../assets/images/teaching/mlprojects/dataviz/colourmaps.png" alt="Colour maps" style="width:100%"/> <figcaption style="text-align: center; font-size: small">Colour maps that are not perceptually linear can greatly distort the information.</figcaption> </figure> ] .right-column[ * Colour can be used to highlight elements of a figure. * It is important to choose the right colour map. * Colour blindness must be taken into account: 8 % of men and 0.4 % of women with Northern European ancestry experience congenital color deficiency. ] .references[ * Rougier et al. (2014). [Ten simple rules for better figures](https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1003833). PLOS Computational Biology. * Crameri et al. (2020). [The misuse of colour in science communication](https://www.nature.com/articles/s41467-020-19160-7). Nature communications. ] --- ## Ten simple rules for better figures ### 7. Do not mislead the reader > "\[If we rely on the automatic settings of your software it is easy to\] inadvertently misle\[a\]d your readers into visually believing something that does not exist in your data". <figure style="text-align: center"> <img src="../../../assets/images/teaching/mlprojects/dataviz/misleading.png" alt="Misleadning plots" style="width:70%"/> <figcaption style="text-align: center; font-size: small"></figcaption> </figure> * Wrong choices also lead to misleading plots: avoid pie charts and 3D charts. * As a rule of thumb, use the simplest possible plot. .references[ * Rougier et al. (2014). [Ten simple rules for better figures](https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1003833). PLOS Computational Biology. ] --- ## Ten simple rules for better figures ### 8. Avoid "chartjunk" > "\[C\]hartjunk may include the use of too many colors, too many labels, gratuitously colored back- grounds, useless grid lines, etc.". <figure style="text-align: center"> <img src="../../../assets/images/teaching/mlprojects/dataviz/chartjunk.png" alt="Chartjunk" style="width:70%"/> <figcaption style="text-align: center; font-size: small"></figcaption> </figure> .references[ * Rougier et al. (2014). [Ten simple rules for better figures](https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1003833). PLOS Computational Biology. ] ??? * Background colour in a plot is generally a bad idea --- ## Ten simple rules for better figures ### 9. Message trumps beauty > "It is important to know \[the standards in each scientific domain\], because they facilitate a more direct comparison". > "However, most of the time, you may need to design a brand-new figure, because there is no standard way of describing your research". <figure style="text-align: center"> <img src="../../../assets/images/teaching/mlprojects/dataviz/uncannyvalley.png" alt="Uncanny Valley" style="width:50%"/> <figcaption style="text-align: center; font-size: small">This figure is an extreme case where the message is particularly clear even if the aesthetic of the figure is questionable.</figcaption> </figure> .references[ * Rougier et al. (2014). [Ten simple rules for better figures](https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1003833). PLOS Computational Biology. ] --- ## Ten simple rules for better figures ### 10. Get the right tool > "Depending on the type of visual you’re trying to create, there is generally a dedicated tool that will do what you’re trying to achieve". * It is important to get familiar with at least one tool for producing reproducible plots. In Python, popular tools are `matplotlib` and `seaborn`. * There exist multiple great open-source options for computer graphics: * Inkscape: professional vector graphics * GIMP: Photo editing * TikZ and PGF: TeX packages for programmatic graphics. * D3.js: interactive data-based graphical forms .references[ * Rougier et al. (2014). [Ten simple rules for better figures](https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1003833). PLOS Computational Biology. ] --- ## Beyond bar plots .context[Bar plots are a very common choice. However, they are rarely a good choice] <br> .highlight1[Many different datasets can lead to the same bar graph.] .center[<img src="../../../assets/images/teaching/mlprojects/dataviz/barplots1.png" alt="Many different datasets can lead to the same bar graph" style="width:90%"/>] .references[ Weissgerber et al. (2015). [Beyond bar and line graphs: time for a new data presentation paradigm](https://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.1002128). PLOS Biology. ] --- ## Beyond bar plots .context[Bar plots are a very common choice. However, they are rarely a good choice] <br> .highlight1[Bar graphs hide information about individuals, suggesting the groups are independent.] .center[<img src="../../../assets/images/teaching/mlprojects/dataviz/barplots2.png" alt="Bar graphs hide information about individuals, suggesting the groups are independent." style="width:90%"/>] .references[ Weissgerber et al. (2015). [Beyond bar and line graphs: time for a new data presentation paradigm](https://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.1002128). PLOS Biology. ] --- ## Beyond bar plots .context[Bar plots are a very common choice. However, they are rarely a good choice] <br> .highlight1[Bar graphs discourage the reader from thinking critically about statistical significance and the authors's interpretation of the data.] .center[<img src="../../../assets/images/teaching/mlprojects/dataviz/barplots3.png" alt="Bar graphs discourage the reader from thinking critically about statistical significance" style="width:55%"/>] .references[ Weissgerber et al. (2015). [Beyond bar and line graphs: time for a new data presentation paradigm](https://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.1002128). PLOS Biology. ] --- ## Beyond bar plots .context[Bar plots are a very common choice. However, they are rarely a good choice] <br> .highlight1[Conclusions] * Avoid bar plots unless you know for sure it is the right choice: they are appropriate only for counts and proportions and when the deviation of the means is very small or zero (rarely is). * Scatterplots are usually a better choice. * If the distribution has few data points, show them all and represent a summary statistic. * If the distribution has too many data points, show the distribution with a curve, a violin plot, a box plot, etc. ??? ClimateGAN as an example .references[ Weissgerber et al. (2015). [Beyond bar and line graphs: time for a new data presentation paradigm](https://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.1002128). PLOS Biology. ] --- ## Beyond bar plots ### Examples .context[An alternative to bar plots is showing the distribution.] .center[<img src="../../../assets/images/teaching/mlprojects/dataviz/climategan_fig5.png" alt="ClimateGAN, Fig. 5" style="width:45%"/>] .center[<img src="../../../assets/images/teaching/mlprojects/dataviz/climategan_fig6.png" alt="ClimateGAN, Fig. 6" style="width:65%"/>] .references[ Schmidt et al. (2021). [ClimateGAN: Raising Climate Change Awareness by Generating Images of Floods](https://arxiv.org/abs/2110.02871). ICLR 2022. ] --- ## The use of colour .context[Colour is very often misused in scientific communication.] <br> .highlight1[Never use the jet colour map. For many reasons. Use maps that are friendly with colour-vision deficiencies] .center[<img src="../../../assets/images/teaching/mlprojects/dataviz/colourblindness.png" alt="Colour blindness" style="width:55%"/>] .references[ Crameri et al. (2020). [The misuse of colour in science communication](https://www.nature.com/articles/s41467-020-19160-7). Nature communications. ] --- ## The use of colour .context[Colour is very often misused in scientific communication.] <br> .highlight1[Choose perceptually uniform colour maps] .center[<img src="../../../assets/images/teaching/mlprojects/dataviz/perceptualuniformity.png" alt="Perceptual uniformity" style="width:55%"/>] .references[ Crameri et al. (2020). [The misuse of colour in science communication](https://www.nature.com/articles/s41467-020-19160-7). Nature communications. ] --- ## The use of colour .context[Colour is very often misused in scientific communication.] <br> .highlight1[Choose the right family of colour maps: sequential, categorical, diverging, etc.] .center[<img src="../../../assets/images/teaching/mlprojects/dataviz/colourmaps_categories.png" alt="Colour maps categories" style="width:45%"/>] .references[ Crameri et al. (2020). [The misuse of colour in science communication](https://www.nature.com/articles/s41467-020-19160-7). Nature communications. ] --- name: title class: title, middle ## IFT 3710/6759 ## Projets (avancés) en apprentissage automatique #### .gray224[6 février 2024 - Session 9] ### .gray224[Visualisation des données] .bigger[.bigger[.highlight1[Questions, doubts, concerns, comments?]]] .center[ <a href="http://www.umontreal.ca/"><img src="../../../assets/images/slides/logos/udem-white.png" alt="Mila" style="height: 6em"></a> ] Alex Hernández-García (he/il/él) .footer[[alexhernandezgarcia.github.io](https://alexhernandezgarcia.github.io/) | [alex.hernandez-garcia@mila.quebec](mailto:alex.hernandez-garcia@mila.quebec)]<br> .footer[[@alexhg@scholar.social](https://scholar.social/@alexhg) [](https://scholar.social/@alexhg) | [@alexhdezgcia](https://twitter.com/alexhdezgcia) [](https://twitter.com/alexhdezgcia)] <