Please disable Adblockers and enable JavaScript for domain CEWebS.cs.univie.ac.at! We have NO ADS, but they may interfere with some of our course material.
Q8 Answers
- Isn't it a very critical question, to order based on derived attributes. If these attributes are not properly chosen, this could mislead the user.
-
- Should the user be informed on how filtering and other reduction techniques had been applied to the data in order to help him recognizing issues or deceptions that have been caused by the filtering?
-
- Is it a good idea to give the user the option to turn off the filtering and watch the real data if he gets suspicous about special patterns?
-
- From the user’s point of view, this approach is very difficult to use: how on earth do they know what numbers to type? After they type, how do they know whether that choice is correct? (p. 300)
-
- From my point of view I don't find this approach that difficult to use. Because the Viewer of the chart should at least know a little about at what he is looking (hopefully the title will help), and giving him a range to the slidebar (for example 1-100) wouldn't be that hard.
-
- Does a boxplot always show the same 5 attributes? So is the boxplot the best way to visualize Figures? (p. 308)
-
- Typically, the user investigates by clicking on points and seeing if the spatial layout implied by the low-dimensional positions of the points seems to properly reflect the high-dimensional space. (p. 319)Aren't we tryring to avoid that? I mean that click and try approach. Sure it's pretty easy to implement, but what about the usability? Wouldn't it be bad if the viewer has to click like a 100 times until he gets the view he wants? Isn't there a better way then clicking through everything?
-
- I am really interested how the dimensionality reduction algorithms work. Is this simply some clustering approach with prior feature selection? Might have to read that "Glimmer" paper that introduces the algorithm.
-
- In general the whole aggregation and reduction process requires a good amount of knowledge of proper statistical estimators / algorithms. Are there libraries that efficiently implement these and then will simply be linked with a new visualization? While I can think of some great frameworks ( R, scipy ) that offer an extensive collection of algorithms I am not that sure whether they would be used as the "backend" in practice ( or at least not R ).
-
- The extension of the 'standard' boxplots with the distribution information is great, unfortunately I have never seen this in the wild. Is there a lot of exchange between the statistics und visualization people? I think that a lot of standard visualization methods in statistics could easily be improved ( and there already are a lot of improved methods because of the Vis people ). But it seems that the statisticians don't adopt these changes in their tools, or at least not very fast.
-
- Reduction by Manipulation: What are the problems regarding this? I might change the data, what can I do to prevent a change in the message of my visualization as weil? e.g. Aggregation, clustering of vector glyphs
-
- How can I asses potential users of my visualization? They might help me to reduce unnecessary items?
-
- SolarPlot: The book is saying that when you generate a solarplot with a bigger radius it get's more detailed, because of the possibility to have more bins of a histogram. I'm just wondering if the readability isn't quite the same before and after increasing the number of bins. I mean, when I reduce the amount of bins, it should be less complex, but the density of information in relation to the space used to display this information is maybe equal.
-
- I'm not quite sure about this, but unless I'm not mistaken, if seen a vase-scatterplot before in a different context. Is this use of vase plots quite sophisticated/used?
-
- While I was looking at Figure 13.6 I was thinking, that this visualization looks pretty artistic. In the description it's saying that it's visualizing a large dataset. I was wondering how long it would take the computer to calculate to provide that visualization. I've completely no idea of which amount of data we're talking, but is this visualization a project I could easily do on a "normal" notebook?
-
- 13.4.1.:interactively changing the bin size in a histogramm. As it is stated the histogramm will look very differently with differnent bin sizes. Isnt that a big problem and not something the user can explore the data without a lot of knowledge?
-
- 13.4.1.: more sophisticated approach to interactively change aggregation is done as a result of higher-level interaction usually based on spatial proximity. Is that only zooming or are there more examples?
-
- 13.4.3.:dimensionality reduction: If you use the approach how do you find a name or description for the new dimensions?
-
- How is the programmer or the user supposed to know when the vis is filtered to much if not in a kind of subjective way? (I mean by that that the two have to sit together and just go trial and error)
-
- I think eliminating attributes to items can be very irritating for a user. When I have an item and want to make a query for a specific attribute which was not thought of in development the buyer of the vis product is forced to rework it which will produce more costs. Wouldn't it be better to filter out attributes instead of eliminating them in the sense of not implementing them?
-
- I don't understand why the boxplot provides an aggregate view?
-
- Is there a limit that the attributes can be filtered to and what is this limit?
-
- While using aggregation, what should we keep in mind to visualize the data genuine, even if some attributes and/or items are merged?
-
- Is there any other way than using a 2D scatterplot for dimensionality reduction of a document collection?
-
- what is the main goal when inspecting scatterplots ofdimensionally reduced data?
-
- why is a scatterplot matrix(SPLOM) is a good choice whe more than twosythetic attributes are created?
-
- why would chaniging the scale of the units in the spatial aggregation lead todifferent results?
-
- Regarding gerrymandering: Is there a rule of thumb on how to draw the geographic boundaries of a data set to represent the actual data best?
-
- It says in the book, that the dimensions of DR data are not correlated, because all dimensions that are correlated get aggregated into one dimension. How does one reduce data with no or only some correlated dimensions or would that make no sense?
-
- Could you give an example of when to use linear and nonlinear methods of MDS?
-
- Aggregation and filtering are often based on similarity measures. As a start, the linear correlation coefficient is obvious, but which other standard measurements are there? Especially in regard to nonlinear correlation and categorical-categorical or categorical-quantitative similarity.
-
- Are there any widely used approaches to interactive dimensionality reduction? E.g letting the user project attribute combination to lower dimensions.
-
- Solar plots seem to have nice spatial properties when dealing with large attribute sets (since they are circular they are more likely to fit onto a rectangular computer screen than a long histogram). Do they have any other advantageous properties that would justify their use? It seems harder to compare bars than on a normal histogram.
-
Letzte Änderung: 07.05.2015, 18:52 | 1232 Worte