cogcomp / apelles

CogComp-nlp demo

Apelles

This is a visualization library designed around cogcomp's NLP datastructures.

Basics

Requirements: NodeJS, NPM should be installed.

To install dependencies: npm install in the root directory to install required NodeJS modules.

There are two use-cases for the system:

1. Demo for visualization of annotations (served over networks)
2. Diff-ing tool for local files containing annotations

Running the demo server

To start the dev server: node server.js

Local comparison and display tool

comparison.js provides a tool for viewing annotations from static files (TextAnnotations serialized as json}, including comparing several different versions of the same annotation of the same text.

Usage for comparing multiple annotation versions:

1. Create two or more folders, representing versions, of annotation JSON files with identical file names.

   Example: prediction/sample.json and gold/sample.json

2. Start server with node comparison.js --port <PORT> <FOLDER#1> <FOLDER#2>. This command can be run from any working directory, with <FOLDER> relative to current working directory.

   Example: node comparison.js --port 3154 prediction gold

3. Browse <SERVER ADDRESS>:<PORT>. All files named according to 1. will be shown under the first drop-down list. Multiple files and view types can be chosen to display vertically in order.

   Example: localhost:3154

Usage for visualizing a single version of annotations:

Same as for multiple versions, but specify a single directory.

Adding a new annotation type

If the name is not already in the available view list, in apelles/public/comparison.html edit the selectpicker element <select class="selectpicker col-xs-12" id="view-selector" data-actions-box="true" multiple> by adding a new option: <option value="MENU_VERSION_OF_YOUR_VIEW_NAME" selected>YOUR_VIEW_NAME</option>

apelles's Issues

Brat to be moved inside the renders?

So about the current structure of the visualizations. The Renderers return blocks of entity/relations/etc. I think they should be more abstract: like they should get json object and name of the block in which they should be shown. In other words, brat’s Utils.embed(.) should be called inside the rendered. Basically the user of the library shouldn’t have to deal with the brat.

Integration with Pipeline Webserver does not work

Annotating new text in the TextBox of the webpage does not call the pipeline server correctly.

Use TokenOffset field from the JSON

The new JSONSerialization write tokenOffsets for each token. We need to use this field to create the raw_text in https://github.com/CogComp/text-annotation-viz/blob/master/index.js#L72

The delay in loading the fonts.

When the page loads, it takes 5 seconds to load the fonts.
Hence it feels that the first query is very slow (or even not working).

Noted by @nitishgupta

Visualizing annotations within files

Suppose I want to visualize annotations within json (annotation) files.
Is there a way to do it? (note that this is not diff-ing, but it should be easier).

@mssammon @lxtbell

@ChaseDuncan this was your question; right?

Many Views are not supported

• PredicateArgument View
• Tree View
• Coreference View

Add a loader

https://www.w3schools.com/howto/howto_css_loader.asp

Comparison not aligned for long and uneven files

Lines in the side by side comparison are not aligned if files are long and contain vastly different number of labels, since a line without labels will take much less height.

Generating the javascript file

Just to remember it:

npm install browserify -g
browserify index.js > apelles.js

Annotation multi-selector unreliable?

Stops working once in a while. Will have to verify its functions.
Details here: davidstutz/bootstrap-multiselect#892

Use Christos's viewer

https://github.com/christos-c/tree-viewer

share-able link

We should have the possibility of having shareable link for a single visualization instance.

Add a limit to the input sentence length.

And show a proper msg upon exceeding it.

Show a proper msg when the server is down is not responding.

In fact it should be checked upon loading the page, as well as submitting any request.