[CT-1091] [Feature] Add Lookback Attribute for Metrics

[callum-mcdata]

Is this your first time submitting a feature request?

• I have read the expectations for open source contributors
• I have searched the existing issues, and I could not find an existing issue for this feature
• I am requesting a straightforward extension of existing dbt functionality, rather than a Big Idea better suited to a discussion

Describe the feature

The metrics package is looking to expand the types of metrics that it can operate on and one consistent piece of feedback we've received is that people want to be able to define metrics that have a concept of time included in their definition. IE 14 Day Average, etc.

This functionality has an open PR here for the metrics package. We've proven that it works but want to move the definition of this metric aspect into the metric itself instead of meta.

Example:

metrics:
  - name: base_lookback_metric
    model: ref('fact_orders')
    label: Total Discount ($)
    timestamp: order_date
    time_grains: [day, week, month]
    type: sum
    sql: discount_total
    dimensions:
      - had_discount
      - order_country
      
    window: 14 days
   

Describe alternatives you've considered

Not including it.

Who will this benefit?

Anyone who wants to use metrics that are cumulative or lookback.

Are you interested in contributing this feature?

Yep!

Anything else?

I am not married or even really attached to the name of lookback. It is a placeholder while we work on the functionality.

[jtcohen6]

Thanks for the issue and the PR @callum-mcdata! Adding support for rolling lookback windows is cool and good. Let's add this into the v1.3 initiative around improvements to metrics.

Very open to suggestions from the world (or just @dbeatty10) about what we'll want to call this thing. I don't mind lookback personally.

[dbeatty10]

Love any opportunity for a good brainstorming session!

Some quick ideas for nomenclature:

lookback
window
offset
lead/lag
start/end
window_bounds

Some of these might be able to be dismissed quickly.

window is probably my favorite so far. Not because I've spent a sufficient amount of time understanding the context though! 😅 Mainly because after some brief googling, learned that window is the only required parameter of Series.rolling() within Pandas (which return Rolling objects that support a variety of aggregation functions). Learn more here.

Example usage:

In[1]: 
s = pd.Series(range(5))

s.rolling(window=2).sum()
Out[2]: 
0    NaN
1    1.0
2    3.0
3    5.0
4    7.0
dtype: float64

[callum-mcdata]

I think @dbeatty10 convinced me that window is a better descriptor of what we're working with.