omgpotatoes / emr-vis-nlp Goto Github PK

the current focus document changes when 1: the focus document belongs to a multirow nested grid, and 2: the number of row items changes in response to the resize

ensure feedback works correctly also

add support for focusing on particular regions of a document, based on user-selected terms, ML weights, etc.

agglomerative clustering?

replace jtable with jtree?

Expand functionality of histogram: highlight relevant documents on hover over bar, adjust % of each bar that is filled when documents are hidden/unhidden. Also, hidden documents should not consume space in the visualization! Rather than simply not rendering the documents, they should also not be included when performing the initial layout.

Currently, documents are assigned grid based on manual annotation, and position based on certainty of prediction. This is OK, but we need to ensure that:

1. documents without a manual annotation are assigned an appropriate cell based on prediction
2. there is some visual indication to distinguish manually-annotated docs versus docs which are positioned based only on predictions (or x-prediction y-annotation, or vice versa)

(because the current dataset has all annotations, this major issue has been obscured )

implement in MLPredictor, or move entirely to a View? Controller?

since same attribute set is used regardless;

Or, should this be stored in MainModel, since this is where the manual annotations reside?

No, an updated attribute value should be pushed separately to the MainModel and MLPredictor, since MLPredictor may use additional context (ie, from highlighted span).

after taking documentation screenshot

base on axes rather than color

Rather than rebuilding whole DocumentGrid, smoothly perform reassignment in-place (this is the proper way to do reassignment anyway, and will be necessary if any animations are to be incorporated). Also, don't jump document off of pointer; keep document as active drag object, simply perform reassignment silently in the background (no immediate .run("init") in DocumentGrid).

right now (by default), a string of (whitespace-separated) terms is treated by the Prefuse SearchQueryBinding class as a disjunctive search; all documents are returned which contain at least one of the terms. We would prefer to have control over the searching; it would be best if we could define conjunctive or disjunctive queries, or literal strings. It is not obvious how to enable this behaviour; we may need to extend SearchQueryBinding.

highlight other docs which contain term? highlight sections of other docs that are relevant to the same issues? (or simplification: other terms that are strongly associated with whatever indicator the hovered term is)

of histogram; (as opposed to current highlighting of single document group.) also support multiple group enabling / disabling at a time.

summaries in each cell?

idea: box around glyphs for single patient? or, patient starts as single glyph, interacting causes it to be split into constituent glyphs representing each report for patient? (constituent docs organized temporally?)

(lower priority)

re-imagine the layout for the DocumentGrid pane as grids-within-grids rather than scatterplots-within-grids; this could help to better organize documents, make the visualization more usable by the target audience, etc.

There is a race condition between appearance of the glasspane vs. disabling the fisheye distortion. Sometimes the fisheye distortion is properly frozen upon loading of the glasspane, while other times it incorrectly resets before the glasspane appears.

also: slow loading of glasspane could be related to this issue?

Currently, text highlighting and the summarization component are bound to a single selected attribute; extend these so that they apply to both of the axes on the attributes rather than only to the selected/color attribute.