cls1 bug hunt

[ttimbers]

Captions not aligning with figure content

(more like improving the figure captions so that they can more easily stand on their own)

• [ ] To be more precise, we should change the figure legend for figure 5.2 from "Scatter plot of concavity versus perimeter with new observation represented as a red diamond." to "Scatter plot of concavity versus perimeter colored by diagnosis with new observation represented as a red diamond." (bolding just to highlight change)
• [ ] To be more precise, we should change the figure legend for figure 5.3 from "Scatter plot of concavity versus perimeter. The new observation is represented as a red diamond with a line to the one nearest neighbor, which has a malignant label." to "Scatter plot of concavity versus perimeter colored by diagnosis. The new observation is represented as a red diamond with a line to the one nearest neighbor, which has a malignant label." (bolding just to highlight change)
• [ ] To be more precise, we should change the figure legend for figure 5.4 from "Scatter plot of concavity versus perimeter. The new observation is represented as a red diamond with a line to the one nearest neighbor, which has a benign label." to "Scatter plot of concavity versus perimeter colored by diagnosis. The new observation is represented as a red diamond with a line to the one nearest neighbor, which has a benign label." (bolding just to highlight change)
• [ ] To be more precise, we should change the figure legend for figure 5.5 from "Scatter plot of concavity versus perimeter with three nearest neighbors." to "Scatter plot of concavity versus perimeter colored by diagnosis. The new observation is represented as a red diamond with lines to the three nearest neighbors, two of which have a benign label and one of which has a malignant label." (bolding just to highlight change)
• [ ] To be more precise, we should change the figure legend for figure 5.6 from "Scatter plot of concavity versus perimeter with new observation represented as a red diamond." to "Scatter plot of concavity versus perimeter with new observation represented as a red diamond." to "Scatter plot of concavity versus perimeter colored by diagnosis with new observation represented as a red diamond." (bolding just to highlight change)
• [ ] To be more precise, we should change the figure legend for figure 5.7 from "Scatter plot of concavity versus perimeter with 5 nearest neighbors circled." to "Scatter plot of concavity versus perimeter colored by diagnosis with new observation represented as a red diamond. The 5 nearest neighbors are circled." (bolding just to highlight change)
• [ ] To be more precise, we should change the figure legend for figure 5.8 from "3D scatter plot of the standardized symmetry, concavity, and perimeter variables. Note that in general we recommend against using 3D visualizations; here we show the data in 3D only to illustrate what higher dimensions and nearest neighbors look like, for learning purposes." to "Three-dimensional (3D) scatter plot of of the standardized symmetry, concavity, and perimeter variables, colored by diagnosis. The new observation is represented as a red diamond with lines to the five nearest neighbors, four of which have a benign label and one of which has a malignant label. Note that in general we recommend against using 3D visualizations; here we show the data in 3D only to illustrate what higher dimensions and nearest neighbors look like, for learning purposes." (bolding just to highlight change)
• [ ] To be more precise, we should change the figure legend for figure 5.9 from "Comparison of K = 3 nearest neighbors with standardized and unstandardized data." to "Comparison of K = 3 nearest neighbors with unstandardized (left) and standardized (right) data. Each plot is a scatter plot of concavity versus perimeter colored by diagnosis. The new observation is represented as a red diamond with lines to the three nearest neighbors. The neighbours are different for the new observation in the unstandardized and standardized data sets because of the difference in scale between the two variables.**" (bolding just to highlight change).
• [ ] To be more precise, we should change the figure legend for figure 5.11 from " Imbalanced data." to "Scatter plot of concavity versus perimeter colored by diagnosis with imbalanced data (where there are many more benign observations compared to malignant ones)." (bolding just to highlight change).
• [ ] To be more precise, we should change the figure legend for figure 5.12 from "Imbalanced data with 7 nearest neighbors to a new observation highlighted." to "Scatter plot of concavity versus perimeter colored by diagnosis with imbalanced data (where there are many more benign observations compared to malignant ones). The new observation is represented as a red diamond with lines to the seven nearest neighbors, four of which have a benign label and three of which has a malignant label." (bolding just to highlight change)
• [ ] To be more precise, we should change the figure legend for figure 5.13 from "Imbalanced data with background color indicating the decision of the classifier and the points represent the labeled data." to "Scatter plot of concavity versus perimeter colored by diagnosis with imbalanced data (where there are many more benign observations compared to malignant ones). The background color indicates the decision of the classifier (with K = 7) and the points represent the labeled data." (bolding just to highlight change)
• [ ] To be more precise, we should change the figure legend for figure 5.14 from "Upsampled data with background color indicating the decision of the classifier." to "Scatter plot of concavity versus perimeter colored by diagnosis with upsampled data. The background color indicates the decision of the classifier (with K = 7) and the points represent the labeled data." (bolding just to highlight change)
• [ ] To be more precise, we should change the figure legend for figure 5.15 from "Scatter plot of smoothness versus area where background color indicates the decision of the classifier." to "Scatter plot of concavity versus perimeter colored by diagnosis. The background color indicates the decision of the classifier (with K = 7) and the points represent the labeled data." (bolding just to highlight change) Question here, is K=7? Should we add anything to this figure caption about this being the output of the entire code workflow that proceeds it?

Outstanding questions about figure captions:

1. for Fig 5.8 we say "standardized" but we don't for the rest of the variables... Should we? I think we should be consistent between all of these, whichever way we lean.
2. should we use little letters (e.g., "a" and "b" for the subplots here so it's easier to reference them? For example in figure 5.9?
3. Should Fig 5.10 become a subplot (i.e., we combine 5.9 and 5.10 into one plot)? This would simplify the figure legends (less repetition needed to precisely explain what is happening) and make the figure caption more readable (it's easy to miss the figure caption for Fig 5.9 right now as it is small and sandwiched tightly between two plots.

[ttimbers]

Poorly formatted figures (size, aspect ratio, legibility, etc)

• [x] In the legends of some figures in this chapter we capitalize "unknown" and in some we do not. We always capitalize "Benign" and "Malignant". I think we need more consistency here.
• [ ] I don't think there is enough contrast between the data points and the background colours in Figures 5.13, 5.14 and 5.15. It is much worse for the orange points compared to the blue. Also, I think even though I can differentiate, it takes more time for my brain to do this than it should. And folks with accessibility issues will have an even harder time...
• [x] I think the font sizes for the graph labels should be the same between Figs 5.9 and 5.10 (standardized vs unstandardized plots), especially if we combine them into one plot with three panels.

[ttimbers]

Anything else you can think of

• [ ] Should we number formulas and reference them in the text? Like we do with tables and figures?
• [ ] Should we round the code output for the Perimeter column (from the code that calculates the distances for 5 neighbors) to match the significant digits shown in table 5.1? Or alternatively, increase the precision of the permiter column in table 5.1 to match the code output? Also, it would be nice if the columns, and their orders were the same between the code output and the table. That would make it much easier to compare and match the two up (we want students to match up the distance columns. We might also want to have the Distance columns share the same column name in the two versions of this table... We then should modify the code to match for the distance calculation in the "More than two explanatory variables" section.
• [ ] We name the code data sets scaled_cancer and unscaled_cancer but then in the text and figures refer to these as standardized and unstandardized... Are we OK with this? Or should we better match these, i.e., rename data sets standardized_cancer and unstandardized_cancer?
• [ ] The upsampled data in Fig 5.14 looks no different than the unbalanced data in Fig 5.13. This is due to overplotting, however this is likely not obvious to the reader. We should point this out in the narrative (and maybe in the figure caption?)