Skip to main content

Common use-cases

In this section, we'll go over some common use-cases that you might encounter when using Raptor.

How can I "join" between two data sources?​

Rather than joining between two data sources, we can use the concept of "dependencies" or "derived features" to achieve the same result.

To do that, we can use the ctx.get_feature() function to request the value of a feature from another data source.

For more information, check out the Derived feature section.

How can I use have aggregations without time windows?​

Unfortunately, Raptor doesn't support aggregations without time windows at the moment. It's in our near roadmap.

As a workaround, you can specify a very big time window(over) with a granularity of the window size.

How can I use a feature that's not a column in my data source?​

Features can be any value that you can calculate from your data source, or from another feature.

How can I calculate the time difference between two events?​

To calculate the time difference between two events, you should assist another derived feature.

For example, let's say we want to calculate the time difference between the purchases of a customer.

@feature(keys=['customer_id'], datasource=CustomerPurchases)
@keep_previous(versions=1, over='1d')
@freshness(max_age='1h', max_stale='1d')
def last_purchase_time(this_row: CustomerPurchases, ctx: FeatureContext):
  return this_row['purchase_time']


@feature(keys=['customer_id'], sourceless_markers_df=CustomerPurchases)
@freshness(max_age='1h', max_stale='1d')
def time_since_last_purchase(_, ctx: FeatureContext):
  last_purchase = ctx.get_feature('last_purchase_time')
  return ctx.this_row['purchase_time'] - last_purchase

How can I use python packages for my features and models?​

You can use any python package that you want, as long as:

  1. It's installed in your local environment.
  2. You instructed Raptor to install it in the production environment using the @runtime decorator.
  3. You imported it inside your function.
@feature(keys=['user_id'], data_source=User)
@freshness(max_age='1h', max_stale='5h')
@runtime(packages=['phonenumbers==8.13.5'], env_name='default')
def is_valid_phone(this_row: User, ctx: FeatureContext):
  import phonenumbers
  p = phonenumbers.parse(this_row['phone'], "US")
  return phonenumbers.is_valid_number(p)

In the example above, we're using the phonenumbers package to validate a phone number.

We're also using the default environment, which comes with Raptor by default. The environmet acts exactly like " virtualenv" in python - it's a separate python environment(bounded to specific python version), that you can install packages in.

Spawning a new environment is usually done by the DevOps team, and it's not something that you should do yourself.