Chris

REVIEWS

Jimi Radabaugh's Review

The concept behind this series of visualizations is absolutely fascinating and I’m impressed with the overall presentation as well. I agree that four is an appropriate number of bins, as more bins would have increased complexity at the expense of clarity. I understand the rationale behind using multiple color schemes, as the overall presentation is quite aesthetically pleasing, but on the other hand I worry that the different color schemes initially suggest differences in the way the three sections are quantified, which turns out not to be the case. I really like the three levels of detail on display here (movements, sonatas, composers), but they are arranged in reverse order of the way I would like to see them (composers, sonatas, movements). It seems like it would make more sense to start with the highest level (composer) on the left and then to dig deeper in a liner fashion as one moves across the page from left to right. However, I think the “Piano key coloring by percentile of use” key should remain on the left side. It would also be interesting to dig deeper into sonatas and movements for Mozart and Chopin, provided that the information is available. Of the additional dimensions you’ve already mentioned that may be worth exploring (simultaneity, sequence, duration), I think that the duration of notes sounds the most promising. Of course, given how unique the underlying concept is, this project could go in any number of interesting directions. Well done, Chris.

Kathryn Faulkner's Review

First of all, this is a great visualization. The concept is really interesting and different, and you pulled off the execution really well. My only critique is with your use of color. I don't mind the variation of color on the page, but it doesn't seem necessary to me. Also, it prevents me from being able to compare the three sections side-by-side without having to constantly refer back to the key. Also, the colors for the upper percentiles of both the yellow and red color schemes look very similar, so at first it looks like you could compare them, but then you realize that the darkest red in the yellow color scheme means something different in the red color scheme. I think it could be simplified with a single color scheme.

As far as improving on the visualization for further iterations, I think it would be interesting to explore visualizing keystrokes over time. I've seen an interesting music note animation, I think by this place: http://www.musanim.com/. Perhaps that is something worth considering. Another bit of data that might be interesting to include in the current visualization is the total number of key strokes for each movement or sonata. Overall, great job!

Dr H's Review

Wonderfully interesting visualization Chris. It clearly engaged the class. Doing something new and innovative was very inspiring. Taking into account music theory would be interesting (this was touched on), i.e. if you compose in key of B flat for instance then there is higher likelihood of certain keys being played. As you noted in talk, it would be interesting to look for outliers, i.e. keys less frequently used, changes in things like dissonance over career (Pam's suggestion), or used commonly but not in "composition key". Looking at this in time, as you suggested (i.e. change of number of keys on pianos can be seen is very interesting). I also like the idea of playing notes over time. I wonder if you tried other visualization (like stacked chart to show number of times hit, but somehow visualized on top of the keyboard keys)? Using the keyboard was great. personally I liked the colors. You could do greyscale for number of hits, but this would conflict with our real world experience of piano keys (which are black and white). So I think using colors is a good choice. It's certainly looks beautiful!

PRESENTATION

Heatmap Étude: Visualizing Classical Piano Music as a Function of the Piano Keyboard

The idea here is to show the "popularity" of piano keys according to a number of variables (by composer, by sonata, by sonata movement) using a color scale, mapped onto the "geography" of today's standard 88-key piano keyboard.

What do you think? Is this viz an A major or an F minor?

Links to mid-term project images

    • UNC Wordpress PDF link

  • Google Drawings Presentation Copy link

CUTT-AD-DDV

Context: Newspaper info-graphic (~ NY Times DNA example) (with different technology, this could be an interactive online newspaper data tool, more like the NY Times Netflix example)

User: Curious/interested news reader

Task: Peruse the data

Data Types:

    • Nominal: Movement, Sonata, Composer, piano key names

    • ~Continuous (Ratio): # of keystrokes

    • Binned/Ordinal: keystroke percentiles (with corresponding color hues/values)

Experimenting

The source for my data is a collection of presumably handmade midi files meant to represent the pieces of music that are involved. These midi files I converted to piano tablature text files via an online midi-to-tablature tool. I then compiled the tablatures into a spreadsheet that comprises my data set.

I tried out various visual approaches for how to map piano key "popularity" onto the keyboard images. For instance, one question was how many bins to assign to a given keyboard. I tried a number of options, ranging from eight to four. A greater quantity of bins presents more detailed data, which could be good, but I judged that even as many as five bins caused the different keyboard instances to resemble each other. I wanted each instance to be relatively unique, so that visually comparing them would be meaningful. In the end, I've opted for four bins.

The scheme I ended up using represents total keystrokes in four unequal "portions," (1-50%, 50-67%, 67-84%, 84-100%) rather than equal portions (i.e., 1-25%, 25-50%, 50-75%, 75-100%). Again, what this does is to focus the visual information on the more distinctive parts of an instance's data. Do we really care that Mozart uses key x at 23% and Chopin uses it at 31%? Probably not: Neither composer uses key x all that frequently. Instead of polluting the visualization with this irrelevant information, I have opted to focus on each composer's "favorite" keys, reserving three of the four differentiating bins for keys hit at greater than the 50th percentile. Roughly speaking, this boosts the contrast.

Lastly, the visualization I've created does not delve very far into chronological data. Each of the three sections is presented in chronological order, but that is all. Significantly, there is no data about things like simultaneity, sequence, or duration of individual musical notes within a given piece of music. All of these are vital to the sound of a piece of music. Likely future work would be to begin incorporating one or more of these factors.

Resources

PROPOSAL

What I want to do:

Visualize the music of several classical composers as a function of action on the keyboard: number of key strokes; duration of keystrokes (other variables?).

Why this is interesting to me:

I don't play music, or read music, but I would like to -- I would particularly like to play the piano. So, I'm interested in demystifying the keyboard. Analyzing it in this way seems like it could help people like myself approach the keyboard in a different light. Also, I'm interested in how digital technology can contribute to learning to play the piano, particularly by incorporating cues (reactive, colored piano keys, for example) onto the keyboard itself. This project is perhaps a half-step in that direction, by combining musical notation and visualization that could include on-keyboard (or keyboard-shaped) graphics.

Scope of project and expected deliverable:

I plan to collect ASCII-encoded sheet music of three to ten classical conductors (depending some on what I can find), and create visualizations of their music as a function of the keyboard. This will probably mean creating visualizations that look like a piano keyboard, encoded with different kinds of information. Ideally this would be an interactive visualization (e.g., clicking on a certain composer shows his/(her?) data, choosing keystroke #s vs note duration, etc. But I don't yet know what technology I would or could use. If nothing easy/good enough exists, this would become a collection of static images.

So far, I'm working alone, so all responsibilities are mine.

MY REVIEWS OF OTHERS' PROJECTS

for Madeline Coven

I looked at the supporting materials (i.e., the project description and the Wikipedia entry for urban transects); in general, this seems like a very interesting and highly promising topic for a visualization. I'm intrigued by the idea of imaginative maps, where the content of a thesis is essentially schematic in nature. This of course gets very close to art. Mapping a theoretical space, such as the urban transect, is a fascinating example of praxis, the putting onto effect of a theory.

I am trying, then, to understand what the visualization here depicts, in terms of the urban transect. I was not present during the presentation of this project -- which may in fact be a plus, because without a revealing soliloquy, I have no exposition to aid me; the visualization must speak for itself. I think that perhaps I am missing some contextual information, which ought maybe to be added to the visualization in order to aid viewers. I'm not sure precisely what the conceptual mapping is that is being used. My first instinct is to assume that the x and y dimensions describe a spatial mapping, but I'm not certain that this is the case -- particularly given the "pavement cracks and other small spaces" block of yellowish color. Clearly, pavement cracks are dispersed throughout an area; they do not congregate in large homogeneous areas equal in size, and adjacent to, a park or meadow. Examination of the visualization seems to indicate that the "mapping" is an abstract one. But I don't understand the terms of that mapping. What do the sizes of the squares indicate? What does the relative proximity of one square to another indicate? What is indicated when one color-block borders on another? Is a relationship implied? These questions obtain for both the upper and the lower portions of the visualization. In brief, I again think that there is some amount of exposition or context that is possibly missing, some explanation that tells me what it is I'm looking at -- essentially, what the design "algorithm" is.

As far as the use of colored squares to impart information, I think this is promising. In particular, the lower portion of the visualization is interesting in this regard, because I perceive that a single plant (which takes up four squares) is being variously categorized as belonging to up to four different groups. As long as you are sure no instance will ever belong to more than four groups, this seems like an effective strategy.

And concerning color choice, you have several colors in play, but without too much interference. Each one is fairly distinct and easily mapped to the legend -- except perhaps for the two red hues in the bottom portion. The "food and habitat for wildlife" and "the invasive plants" colors are, at least in the digitized version on the PowerPoint slide, difficult to distinguish. It would probably be wise to find another hue for one of these categories. You might also consider duplicating the yellowish "pavement cracks and other small spaces" legend element in the bottom portion, to make sure it is clear that this is what the yellow indicates (I assume it in fact does indicate this?).

for Eddie Moss

Eddie, I like what you have here. I notice you came to a similar conclusion to me: it is possible, at least theoretically, to class repeated small multiples as a kind of static version of an interactive interface. You can kind of think of it as similar to the way cartoonists animate movies, i.e., one frame at a time. So I think this kind of project is useful in working towards making an interactive tool.

I see what you mean when it comes to the bar graphs as being superior to the map and the scatter plot. In fact, the scatter plot ends up almost acting as a fuzzy bar graph -- although, it does give you a sense of the outliers, which could be nice in certain situations. But I think that those situations are likely to be deeper analytical ones, as opposed to the rougher, quicker ones you imagine your tool being used for. To me scatterplots are more for expert use, for people who understand fit lines and statistics, which is probably not most people who might want to look at the data you are visualizing, so I think your bar graphs are for the most part a good choice.

I think I disagree with you, though, about jettisoning the line graph outright. I think you're right in some, maybe most cases. For example, the first bar graph, labeled Destination, with Newark on top and Denver at the bottom. It is true that these entities are not related in a manner appropriate for a line graph. In this case, the main comparison is between different nominal data points, i.e., between cities -- and here they are ordered by magnitude, which would write a single upward slope, which does not a helpful line graph make. But what about the one below it? The overall average by date? Here the comparison is between time periods, and it seems to me that this graph could be (I won't argue that it has to be) made into a line graph. This is the traditional use for a line graph, isn't it? The same conceptual mapping would be made for things like average temperture by month for a city, or New York Stock Exchange data over time -- what's being mapped is a trend.

I also differ with you in the choice to toss out distance information. I beleive that there are factors correlated to distance that might be interesting here. Perhaps you might find distance corresponding to adjusted arrival delays, but not to departure delays -- because the longer you're in the air, the more likely it is that adverse weather, etc., could delay you. This might indicate that the airline should focus its efforts on departure delays, over which perhaps they can exert more control than they can exert over the weather (although of course, preparing for longer flights might concievably take longer than preparing for shorter ones, so this could muddy the interpretation of the data).

And lastly, I would suggest you think about reconsidering your use of color. It strikes me as at best redundant, and at worst distracting, to always highlight the bar with the greatest magnitude by giving it a distinct color (ie., red here). In the horizontal bar charts, the greatest magnitude is indicated by a) the bar's length, b) its position at the top, and c) its color. This strikes me as overkill. I think that the longest length is sufficient. Even if you re-ordered the chart according to some criteria, you'd still be able to make sense out of it by appealing to the relative lengths of the bars, without the use of the red color. Then, if you are going to have a highlighted bar for details-on-demand, just highlight that one (as opposed to having competing highlighting schemes, such as red vs. black-outline). It's just an opinion. But maybe somebody would want to focus on something other than the greatest magnitude.