Performance is better with server = false, which should not be that way

Table is already exportable. This will probably be put on the back-burner as you can still save the graphics by left clicking and save image. Unfortunately Shiny does not have a very good built in download system, so likely would have to build our own. I suggest maybe doing it via an HTML5 canvas.

this error too on the "outlier" tab

Fix this later

alternately, enable loading from url, and in docs, provide url of some example csv files.
note that we can store them at http://openconnecto.me/data/public/

probably makes sense to do all the computations, and then send them an email that has a link to the results, which are still interactive, but maybe a bit less (ie, changing which points are outliers will be hard to do interactively).
this means we'll have to make some default choices, but i think it will be awesome.
i used it yesterday with 1000 points & 10 features and it was ok, but getting slower.

of note, the people i demo'ed it to LOVED it.

is often useful

what do you think about shortening tab names?
in particular, they don't all fit on a row for me.
i propose:

• upload
• heatmap
• sample summary
• feature summary
• outliers
• correlation
• embedding & clustering

Error is in computation of mahalanobis distances. The built in R function uses matrix inversion. See http://stats.stackexchange.com/questions/41220/mahalanobis-distance-on-singular-data.

Sample images:
Sample.pdf

Likely because htmltools hides my_data from shiny

iris.txt

not saying we should include this, but it is a neat idea. and might be useful.
Hubert08a.pdf

upon trying to upload this csv file
smallSynapse.txt

For example, can we plot marginal distributions conditioned on class labels (i.e., grouped by an arbitrary column)

it might look like this:
https://chrisjmccormick.files.wordpress.com/2014/08/1d_example.png

I.e. for 1 million rows it requires 1 Tb of ram.

Not really sure how to deal with this. This might be a good starting point:
http://stackoverflow.com/questions/13867325/get-width-of-plot-area-in-ggplot2

• Option to export plots
• Check R score for scatterplot, etc.
• on "Data" tab, show: # features, # samples, nnz, # NaN
• on "Marginal" tab, make "combined" the default.
• throughout, rename tabs to be more informative
• "Distance Metric" = "Euclidean Distance"?
• fix size of correlation matrix
• mean vector plotting doesn't really have scatter, probably we want 'jittered scatter'
• add iris.csv file for demo purposes, you can leave the other one, maybe name it something more informative.
• Implement robust distance metrics
• tell people the appropriate format (if nan's/infs break code, rows should be data points, columns should be features (or vice versa), appropriate size for things to work)
• Something like scree for tsne
• Better caching for tsne
• Find std of R-scores
• Allow for boxplot scaling
• Consider using scalefillbrewer(type = 'div') for heatmaps
• Scree plots look wrong
• Add kmeans++ clustering
• Check if outliers removed from embed
• Color scheme setting each page?
• cluster on embedded points.
• choose # of dimensions via Zhu & Godsi: https://uwaterloo.ca/data-science/sites/ca.data-science/files/uploads/files/automatic-dimensionality-selection-from-scree-plot-use-of-profile-likelihood.pdf (getElbow.R lives in here)
• we can later enable user to choose the dimension of embedding.
• matrix completion
• Add option to replace Nas Infs?
• Have a link back to issues in vector explorer

heatmap

• for the heatmap, we should enable setting the number of bins to actually discretize the data that way, and use it throughout.
• when i click on stuff, the row always says 1
• we only need to show 2, maybe 3 sig digits, and we should always show the same number.

Summary Sample

• when conditioning on class, we should show class specific means & medians

Outliers

• "Color based on class?" actually changes shape based on class. color based on class might be more clear?

Other

• removing outliers does not seem to work.

@jovo

May want to allow for single column selection? The issue is that in conversion from dataframe to vector, the column names are lost.