How Icicle Makes Collaboration Easy

By allowing queries to be specified separately in an intuitive module system, Icicle provides a great collaborative experience.

To see this in action, let’s imagine we’re a large media organisation, and want to look at how our customers are exploring our sites.

We’ll start with a simple model for a customer’s interaction with a page. For this we define a record named page_views as input stream

input page_views : {
    url : String
  , referrer: Option String
  , category: String
  , subcategory: String
}

This type represents our raw data; and all input streams carry two implicit pieces of information:

• A primary key (think user), and
• A time at which the event was recorded.

With this we could imagine that two data scientists may want to know: have any of the user’s last 10 pages views been to the home screen; and what’s the number of views for sports in the past 4 weeks.

They might write this as:

feature viewed_homepage =
  from page_views
    in latest 10
    in any (category == "home")

feature mean_view_time =
  from page_views
    in windowed 4 weeks
    in filter category == "sport"
    in count category

To introduce a new query, no existing features need to be touched, which makes reviewing new queries easy.

When a large number of queries are specified, it can make sense to break them up into separate modules, and Icicle makes this easy. One module can import functions and input definitions from others.

If one finds that that they are reusing similar terms in their query expressions, they can reduce repetition by writing functions, aiding long term maintenance.