CompassQL is a visualization query language that powers chart specifications and recommendations in Voyager 2.
As described in our vision paper and Voyager 2 paper, a CompassQL query is a JSON object that contains the following components:
-
Specification (
spec) for describing a collection of queried visualizations. Thisspec's syntax follows a structure similar to Vega-Lite's single view specification. However,specin CompassQL can have enumeration specifiers (or wildcards) describing properties that can be enumerated.1 -
Grouping/Nesting method names (
groupByandnest) for grouping queried visualizations into groups or hierarchical groups. -
Ranking method names (
orderByandchooseBy) for ordering queried visualizations or choose a top visualization from the collection. -
Config (
config) for customizing query parameters.
Internally, CompassQL engine contains a collection of constraints for enumerating a set of candidate visualizations based on the input specification, and methods for grouping and ranking visualization.
For example, the following CompassQL query has one wildcard for the mark property. The system will automatically generate different marks and choose the top visual encodings based on the effectiveness score.
{
"spec": {
"data": {"url": "data/cars.json"},
"mark": "?",
"encodings": [
{
"channel": "x",
"aggregate": "mean",
"field": "Horsepower",
"type": "quantitative"
},{
"channel": "y",
"field": "Cylinders",
"type": "ordinal"
}
]
},
"chooseBy": "effectiveness"
}
The examples/specs directory contains a number of example CompassQL queries.
To understand more about the structure of a CompassQL Query, look at the Query interface declaration.
- A query's
specproperty implementsSpecQueryinterface, which follows the same structure as Vega-Lite'sUnitSpec(single view specification) but most ofSpecQuery's properties have-Querysuffixes to hint that its instance is a query that can contain wildcards to describe a collection of specifications. - Since multiple encoding channels can be a wildcard, the
encodingobject in Vega-Lite is flatten asencodingswhich is an array of Encoding in CompassQL'sspec.
Given a row-based array of data object, here are the steps to use CompassQL:
- Specify a query config (or use an empty object to use the default configs)
var opt = {}; // Use all default query configsFor all query configuration properties, see src/config.ts.
- Build a data schema.
var schema = cql.schema.build(data);The data property is a row-based array of data objects where each object represents a row in the data table (e.g., [{"a": 1, "b":2}, {"a": 2, "b": 3}]).
You can reuse the same schema for querying the same dataset multiple times.
- Specify a query. For example, this is a query for automatically selecting a mark:
var query = {
spec: {
data: { url: "node_modules/vega-datasets/data/cars.json" },
mark: "?",
encodings: [
{
channel: "x",
aggregate: "mean",
field: "Horsepower",
type: "quantitative",
},
{
channel: "y",
field: "Cylinders",
type: "ordinal",
},
],
},
chooseBy: "effectiveness",
};- Execute a CompassQL
query.
var output = cql.recommend(query, schema);
var result = output.result; // recommendation resultThe result object is an instance of SpecQueryModelGroup (ResultGroup<SpecQueryModel>), which is a root of the output ordered tree. Its items property can be either an array of SpecQueryModel or an array of SpecQueryModelGroup (for hierarchical groupings).
The SpecQueryModel is an class instance of a SpecQuery with helper methods. Note that, in the result, all of spec query models are completely enumerated and there would be no wildcard left.
- Convert instances of
SpecQueryModelin the tree, usingSpecQueryModel'stoSpec()class method and themapLeavesmethod.
var vlTree = cql.result.mapLeaves(result, function (item) {
return item.toSpec();
});- Now you can use the result. In this case, the tree has only 2 levels (the root and leaves). We can just get the top visualization by accessing the 0-th item.
For a full source code, please see index.html.
var topVlSpec = vlTree.items[0];-
The root file of our project is
src/cql.ts, which defines the top-level namespacecqlfor the compiled files. Other files undersrc/reflect namespace structure. All methods forcql.xxxwill be in eithersrc/xxx.tsorsrc/xxx/xxx.ts. For example,cql.util.*methods are insrc/util.ts,cql.queryis insrc/query/query.ts. -
TODO: constraints
- List in Vy2 paper supplement..
You can install dependencies with:
yarn installYou can use the following npm commands such as
npm run build
npm run lint
npm run test
npm run cover // see test coverage (see coverage/lcov-report/index.html)
npm run watch // watcher that build, lint, and test
npm run test-debug // useful for debugging unit-test with vscode
npm run clean // useful for wiping out js files that's created from other branch
(See package.json for Full list of commands.)
To play with latest CompassQL in the vega-editor, use branch cql-vl3 in kanitw's fork, which has been updated to use Vega-Lite 3, Vega 5, and CompassQL ^0.21.1. (For CompassQL 0.7 or older, use branch compassql, which uses Vega-Lite 1.x).
Make sure to link CompassQL to the editor
cd COMPASSQL_DIR
npm link
cd VEGA_EDITOR_DIR
npm run vendor -- -l compassql
(You might want to link your local version of Vega-Lite as well.)
The main method is cql.recommend, which is in src/recommend.ts.
examples- Example CompassQL queriesexamples/specs– All JSON files for CompassQL queriesexamples/cql-examples.json- A json files listing all CompasssQL examples that should be shown in Vega-editor.
src/- Main source code directory.src/cql.tsis the root file for CompassQL codebase that exports the globalcqlobject. Other files undersrc/reflect namespace structure.- All interface for CompassQL syntax should be declared at the top-level of the
src/folder.
test/- Code for unit testing.test's structure reflectssrc's' directory structure. For example,test/constraint/test files insidesrc/constraint/.typings/- TypeScript typing declaration for dependencies. Some of them are downloaded from the TypeStrong community.
- When you add a new source file to the project, don't forget to add the file to
filesintsconfig.json.