top of page
Lumia Learning - Data
NOTE: Lumia Learning does not capture any camera footage or data other than to track when a person enters the space in front of it and the length to which they stay. Each person that enters Lumia Learning's view is anonymized and no recognition beyond body presence / skeletal tracking is utilized.
Final Report - September 1st
Lumia Learning ends its run at the Del Amo Fashion Center in Torrance, California tomorrow. As part of Torrance Art Museum's Ultra! public art exhibition, Lumia Learning had a unique opportunity to exhibit in the commercial mall setting - a setting, I might add, that it was not necessarily originally intended to be situated (intended at first for a more gallery exhibition environment.) This caused a few wrinkles in Lumia Learning's gathering capabilities - namely that traffic was nigh constant, with people standing in front of the piece at all times of the hour, at all hours of the working day. Lumia Learning was meant to probe when people stood in front of the piece, and for how long, and to tailor itself to the preferences of that viewership. How could it do so, when no clear preferences occurred due to the constancy of its view? What resulted then is a demonstration of the concepts behind the piece taken to the extreme, but resulting in statistics more minute. Still, the data lays out some interesting findings...
Over the course of Lumia Learning's run, despite the constancy of traffic, the piece was able to discern a slight preference for blue colors. As a result, Lumia Learning slowly... ever so slowly... tipped in that direction (slow due to the pull of data showing still consistent preferences for the other colors as well.) See below animation for how the color changed during the course of Lumia Learning's run. This comprises data from July 17th to September 1st, and the way that Lumia Learning changed its colors to suit audience viewership.
On a longer timeline, one can see how blue would remain dominant to eventually overtake the other colors. Again, however, the drag provided by still viewership of the other colors prolonged this process in the high traffic environment of the Del Amo Fashion Center, and still, even at its conclusion for this exhibition, Lumia Learning maintains a wide spectrum of colors shown.
All through it's run, Lumia Learning observed a preference for faster and faster speeds of the moving shapes of the piece. To react, Lumia Learning incremented the speed faster and faster in response. That is, until the last week of its run, when finally, the tolerance seem to even out. Here is every graph in previous reports for speed. Note that the more red the color, the faster the speed.
Though slower speeds provide some preference and drag on the fast speeds becoming dominant, one can see that faster held an edge. Compare these graphs to the last week of week or so of Lumia Learning's run, however:
Clearly the preference for faster speeds found its limit, as the piece settled on a speed that, for reference, was three times as fast as when the piece began, but capped out to not go beyond. One wonders, in the environment of a commercial shopping center, if faster motions and speeds were simply more attracting of attention. What would be observed in a gallery instead?
The last main variable able to be examined in this run was difference in choice between the color changes of the foreground shapes as opposed to the background. Lumia Learning test for linear ramps and falls as opposed to a more curved, parabolic change. It then reversed these, testing for rise and fall vs. fall and rise. In the previous report, it was noted that though the preferences for the different ramps were close, but that a linear ramp rise-and-fall held the edge.
As-of-yet unseen were the results from a linear fall-and-rise. Now, as Lumia Learning wraps up its showing for this exhibition, the results are in:
Linear fall-and-rise proved to be extraordinarily unpopular - less viewed than the other ramps by nearly half. Now, the caveat here: ramp was tested week over week, unlike other variables which followed an hour by hour, twenty-minute-loop cycle. The representations in this graph could therefore easily by explained by external factors. How does the delta-variant or, say, back-to-school schedules affect the attendance of the Del Amo Fashion Center, and therefore, viewership of the piece itself. In future revisions, Hue Ramp will be tightened to test on a less long-term timeline so as to avoid the natural ebbs and flows of attendance -- and to test for percentage of viewers preferring instead of number. In this run, however, Linear ramp rise-and-fall wins out.
As Lumia Learning finishes its run at the Del Amo Fashion Center it has gone through some interesting changes. It is significantly more blue than when it began, running at a faster speed, and changing the colors of the shapes within in a hue shift from one side to the other in a linear rise and fall. It will be fascinating in the future, should Lumia Learning exhibit in a different environment to compare these results to those and see what changes can be observed. For now, Lumia Learning runs for one more day. If in Torrance, California, please, take this opportunity to go see the piece. Special thanks again to the Torrance Art Museum for exhibiting Lumia Learning in their Ultra! exhibition, and to Discover Torrance for housing the piece. Stay tuned to this website for exhibition runs of Lumia Learning in the future.
Report 5 - Tuesday, August 17th
Lumia Learning is nearing the completion of its time at Del Amo Fashion Center, and continues to refine itself to the tastes observed while there. The consistent traffic of the mall has altered the piece's intentions a bit, as preference was designed to measure spikes and valleys instead of plateaus. Still, the piece has adapted in some interesting ways...
Blue continues to weigh the scale heavily, with orange all but disappearing in the statistics. This has caused Lumia Learning to omit orange entirely from future runs. There will still be a brief orange transition from yellow to red, but Lumia Learning will not consider this a true orange representation, and will instead fall on either side of the color. Meanwhile, blue will continue to crawl to dominance, and by the end of the piece's run, I suspect Lumia Learning will be shades of blue semi-exclusively.
Speed In previous reports, it was noted that faster speeds were preferred more and more, pushing the piece to move at a higher rate. Now, we're seeing a curious equilibrium, wherein faster speeds and slower speeds are becoming equally preferred - while oddly not favoring the gap between. Lumia Learning's mode in this situation is to wait and see.
In the last report on Hue Ramp, a shocking bit of data was revealed indicating almost exact preference between linear and parabolic ramps. Again, when seeing this sort of middling and non-preference, Lumia Learning falls into a holding pattern. Fortunately, here, we have the ability to test for more incremental variants - ramping high-to-low, for instance, instead of low-to-high. In the time since, Lumia Learning has tested three ramp variants across nine sequences: Parabolic down-then-up, Parabolic up-then-down, and Linear up-then-down. In these tests, Linear is showing a preference, while the parabolic ramps fall back slightly. Lumia Learning still has to test Linear down-then-up. If it shows a clear preference, Lumia Learning will favor one of the two and move to testing some more detailed sequences within the ramps. 11 days until we see what the final form of the Del Amo Fashion center exhibiting brings...
Report 4 - Thursday, August 5th
When this piece started to exhibit, I, like a poor scientist, made some assumptions and predictions for what the data was going to show. I off-handedly predicted to the people around me as I installed that Lumia Learning would probably not favor blue colors, as they were too close to a default video screen -- or worse to a catastrophic computer error "blue screen of death." I had also predicted slower speeds would likely be the more compelling (this, based on my preference for the calming motions of the work.) As I have seen over the past week or so, my predictions were not just incorrect... but incorrect by a mile...
In the last color report, it was noted that blue held a slight edge over other colors and the piece was beginning to favor them in display. Since that time, the feedback loop of favoring blue, thus it being shown more, thus people seeing it more has caused the blue colors to carve out a heavy dominance in display. Oddly, an accompanying phenomenon is that reds, which appear just before the blues, are also taking a slight edge. This has yellows and oranges fading backward while the other side of the spectrum takes dominance. In the future, the piece will favor blues and, to a less degree reds, while omitting oranges, and beginning to omit yellows and greens.
Even after faster speeds showed a preference, I assumed that when the piece increased the speed of motion, there would be a middling effect wherein the high end was too fast for viewers. So far, that just isn't happening and the refrain of "faster, faster" calls out from the data observed. Lumia Learning will continue to increase its speed, favoring a faster speed of motion. Care must be taken here, however, as too fast could create a strobe-like effect that could be dangerous. I will be monitoing Lumia Learning's speed closely to make sure the work does not become a danger. Commentary here on the preference for blinking, flashy moving objects, particularly in a commercial environment, I will leave to others to make...
In keeping with the theme of today's report, I had assumed that when comparing different days over days, weeks over weeks of viewers, that the piece would be able to make clear differentiations in preference between this and that -- between a, b, or c. Hue Ramp testing by Lumia Learning has once again put my predictions to shame.
In a shocking bit of data, the preference for viewing a parabolic curve for hue ramp-and-fall has proven nearly exactly identical to the views of a linear curve. I can't stress enough that this is over a long period, with many, many views recorded. One would assume that the numbers would offset even just a little -- if not simply due to mall attendance rising and falling one week to the next or one day to the other. Instead, the data is startling consistent with the coin refusing to flip. Lumia Learning is built to observe preference and favor that leaning. It assumes that viewers will inherently be drawn to one set of forms over another in pursuit of an ideal aesthetic. What if this entire core concept is flawed? What if, and particularly in commercial environs, there is no preference to be indicated? An interesting follow-up to this data will be to exhibit in a gallery environment and see how the data observed swings differently than in the Del Amo Fashion Center. Until then, we can only surmise why Lumia Learning remains stuck, testing between two hue ramps. Lumia Learning is currently testing inverted ramps (fall and rise vs. rise and fall) to ascertain if there is a preference there instead. Any poor scientist betters want to make some predictions with me?
Report 3 - Monday, July 26th
Right now, I use a variety of remote tools to check on Lumia Learning, making sure it's functioning without having to physically be in proximity of the piece. At its best this allows me to seamlessly make sure that everything is working properly, and to watch the live data roll in... at its worst it can feel a bit like observing a rover on an alien planet with all the quirks and small errors and glitches that entails. Still, the experiment and piece continue and fighting errors simply becomes routine. All that aside, after being pushed and pulled back to status quo over the past week, Lumia Learning is starting to undergo some changes...
In the previous two data reports, I noted that Lumia Learning was seeing viewers approaching so frequently that there was little room for any one color to stand out. As soon as one did so, it would be cancelled out by the equal pull over the next loop iterations in the opposite direction. This meant that in terms of color, Lumia Learning was stagnant - simply doing a full spectrum color sweep repeatedly. However, this weekend, Lumia Learning observed enough of a color preference that it is starting to tip in one direction. Initially, I admit that I was skeptical of the changes, as, when looking at the graphed data it does not appear that any preference is being indicated...
Grouping those colors together into broader spectrum values, however, reveals the impetus for shifting.
(For added clarity - prior to the weekend observations, those groups were all at 20% equally.) We can see that blue colors are observed 4% more often than orange colors as the piece cycles through the spectrum, causing Lumia Learning to lean in this direction. This preference has previously been so slight that it was easily counter-balanced. Now, however, as the piece errs on the side of that color, it's safe to predict that the feedback loop will cause blues to be preferred as Lumia Learning shows them at a higher rate than the others. Or not! We'll see. For now, blue colors will appear more often and for longer as the piece tips that direction.
Lumia Learning is still observing a preference for faster speeds in the moving shapes and continues to incrementally ramp them up. The 'drag' on this data is caused by the fact that slower speeds are still somewhat popular - faster speeds simply more so. Still, the data bears out that faster speeds are preferred and the speedometer slowly ticks up for speed of motion within the piece.
While color and speed have been easily observed (albeit slow to change), hue ramp testing has been harder to quantify. For reference, the moving shapes within Lumia Learning operate on both a color spectrum sweep and a linear ramp-and-fall for hue. This, in an effort to prevent the colors from being the same on background and foreground, and thus, becoming just a static color. Beginning today, Lumia Learning will be testing to see if there is preference in how this hue ramps and falls, moving between a linear ramp like this...
And a parabolic ramp like this...
...while testing alternative swaps (Ramp Down/Up, Parabolic Up) for preference between. Once the piece has a clear tipping point toward one, it will begin offsetting this against the background color to try to reach an ideal. Time will tell where Lumia Learning takes that data from here. For now, it sits in the planet Del Amo Fashion Center, observing the fauna and attempting to learn.
Report 2 - Wednesday, July 21st
In the first bits of data coming from Lumia Learning, we saw that color choice was inconclusive - with so many people viewing the piece over the course of its twenty minute loops, that there was very little opportunity for the piece *not* to be tracking a person. This was especially true of opening day - a Saturday and the opening of the Ultra! exhibition - in which the numbers of viewers trended so much higher than on a weekday. So much so that those weekend viewers are bearing heavy weight on the data going forward. For example, Lumia Learning has observed that light green was conclusively and by far the most observed color on Tuesday. What the data obscures however, is how few individuals observed the piece that day as compared to the weekend. Which is how we get from a Tuesday color preference chart like this...
To an overall color preference chart like this...
(Forgive the styling differences. The first is the raw data that Lumia Learning spits out, the second is my cleaned up ongoing dashboard.) Here, we see that the great middling effect I spoke of in the text surrounding the piece is coming to fruition. A broad enough observation of participants results in a lack of consensus and a middling of nearly everything being preferred equally - despite clear preferences being shown on a smaller scale. While green, light green, mustard yellow, tangerine, deep orange, and indigo hold slight edges above their neighboring colors, they aren't enough to push or pull the color in any one direction as of yet, as they tend to counter each other out. I suspect that, much like the crowds at your local movie theater, weekend viewers will be the element that provides for more decisive sway. That, or, the piece will continue to shift ever so slightly and come to rest in the middle of the spectrum where it began...
The speed of motion of the piece, on the other hand, continues to hold out trends that faster motion is preferred. The piece is already changing direction, but as with color, is tempered slightly in that weekday viewership simply isn't producing large swings in the data. This means that a faster preference observed Saturday and Sunday holds out by sheer number, not by the individual preferences for each day. As a result, and to try to filter some noise out of the data observed, Lumia Learning will reverse the order of the loops so that the faster speeds play in the first loop, rather than the last. This might produce an interesting result - as in, maybe people like to look at Lumia Learning at the bottom of the hour for other reasons. Or, maybe it won't! Again, weekends will likely be paramount in producing shifts, and we will wait and see. For now, Lumia Learning trends slightly faster, and will favor not only that speed moving forward, but will prioritize it in order of playthrough.
Report 1 - July 17th - Opening Day
Inconclusive findings on color. Lumia Learning has observed that while a deep yellow orange tangerine is the most observed color, it is still very closely followed by a more mustard yellow, green, and violet.
Numbers on colors are still very even across the board, and likely more data will be required to push the piece to noticeable difference.
While color preference may be inconclusive, preferences in speed show a more drastic trend. Lumia Learning currently cycles through three speeds over the course of an hour. Speed 3 showed drastic preference during yesterday's observation, garnering viewers at a rate of nearly 2 to 1. As a result, Lumia Learning will move its shapes faster and test for a higher threshold of speed preference going forward.
The moving shapes operate currently on a linear hue ramp. This will be tested on a day against day basis. Results will arrive in the coming days about the preference of hue ramp.
bottom of page