8. Create lookup feature

Create Lookup features¶

We've learned how to define a target and materialize data using observation tables. Now, let's dive into basic feature engineering.

The most straightforward features we can craft with FeatureByte are known as lookup features. These are either direct columns taken from the source table or simple computations that don't require any aggregations.

In [1]:

import featurebyte as fb

# Set your profile to the tutorial environment
fb.use_profile("tutorial")

catalog_name = "Grocery Dataset Tutorial"
catalog = fb.Catalog.activate(catalog_name)

import featurebyte as fb # Set your profile to the tutorial environment fb.use_profile("tutorial") catalog_name = "Grocery Dataset Tutorial" catalog = fb.Catalog.activate(catalog_name)

15:30:03 | INFO     | SDK version: 1.0.2.dev46
15:30:03 | INFO     | No catalog activated.
15:30:03 | INFO     | Using profile: tutorial
15:30:03 | INFO     | Using configuration file at: /Users/gxav/.featurebyte/config.yaml
15:30:03 | INFO     | Active profile: tutorial (https://tutorials.featurebyte.com/api/v1)
15:30:03 | INFO     | SDK version: 1.0.2.dev46
15:30:03 | INFO     | No catalog activated.
15:30:03 | INFO     | Catalog activated: Grocery Dataset Tutorial

In [2]:

# Get view from GROCERYCUSTOMER scd table.
grocerycustomer_view = catalog.get_view("GROCERYCUSTOMER")

# Get view from GROCERYCUSTOMER scd table. grocerycustomer_view = catalog.get_view("GROCERYCUSTOMER")

Create Lookup feature¶

In [3]:

# Create lookup feature from DateOfBirth column for customer entity.
customer_dateofbirth = grocerycustomer_view["DateOfBirth"].as_feature("CUSTOMER_DateOfBirth")

# Create lookup feature from DateOfBirth column for customer entity. customer_dateofbirth = grocerycustomer_view["DateOfBirth"].as_feature("CUSTOMER_DateOfBirth")

Derive Age at the point-in-time of the request observation¶

In [4]:

# Derive Age from the point-in-time and the date of birth.
customer_age = (
    (fb.RequestColumn.point_in_time() - customer_dateofbirth).dt.day / 365.25
).floor()
# Name feature
customer_age.name = "CUSTOMER_Age"

# Derive Age from the point-in-time and the date of birth. customer_age = ( (fb.RequestColumn.point_in_time() - customer_dateofbirth).dt.day / 365.25 ).floor() # Name feature customer_age.name = "CUSTOMER_Age"

In [5]:

# Transform age into a 5 year age band.
customer_age_band = (
    ((customer_age + 1) / 5).ceil() - 1
) * 5
customer_age_band = (
    customer_age_band.astype(str)
     + "-" + (customer_age_band + 4).astype(str)
)
# Name feature
customer_age_band.name = "CUSTOMER_Age_band"

# Transform age into a 5 year age band. customer_age_band = ( ((customer_age + 1) / 5).ceil() - 1 ) * 5 customer_age_band = ( customer_age_band.astype(str) + "-" + (customer_age_band + 4).astype(str) ) # Name feature customer_age_band.name = "CUSTOMER_Age_band"

Preview feature¶

We will use observation table we created in previous tutorial here.

In [6]:

# Check the primary entity of the feature'
customer_age.primary_entity

# Check the primary entity of the feature' customer_age.primary_entity

Out[6]:

[<featurebyte.api.entity.Entity at 0x176de5c80>
 {
   'name': 'customer',
   'created_at': '2024-04-26T07:28:15.128000',
   'updated_at': '2024-04-26T07:28:24.021000',
   'description': None,
   'serving_names': [
     'GROCERYCUSTOMERGUID'
   ],
   'catalog_name': 'Grocery Dataset Tutorial'
 }]

In [7]:

# Get observation table: 'Preview Table with 10 items'
preview_table = catalog.get_observation_table("Preview Table with 10 items")

# Get observation table: 'Preview Table with 10 items' preview_table = catalog.get_observation_table("Preview Table with 10 items")

In [8]:

# Preview CUSTOMER_Age
customer_age.preview(preview_table)

# Preview CUSTOMER_Age customer_age.preview(preview_table)

Out[8]:

	POINT_IN_TIME	GROCERYINVOICEITEMGUID	CUSTOMER_Age
0	2022-11-03 14:39:00	9faf5936-d4bb-4709-a530-c1624ec003a5	52
1	2023-02-15 21:10:22	5daf8edb-8625-4653-aaa0-9ac03df92017	71
2	2022-10-05 14:30:03	fc4769cf-41d6-4fb0-9cfc-fe30502cfa18	39
3	2022-11-09 18:07:20	37122b82-478b-4d9e-b236-69629b592c0b	27
4	2022-11-18 10:07:49	ae2ccf38-4e5d-4c76-b1e5-04f12307e45b	60
5	2023-03-19 15:22:18	dcdb7c98-dd62-4287-b432-bfe3a2317ebc	22
6	2022-08-16 15:37:21	6781acc6-652d-4867-a138-2d8adb278886	62
7	2023-04-19 15:25:05	8a60f455-aff1-4b4e-8c63-9ab89df2715d	71
8	2023-02-27 13:27:47	b807e05c-ff1c-4fb3-a760-e0e8ce29c859	37
9	2023-03-13 21:11:43	cfc620a2-0054-4b4b-99d5-040cf87cfe2d	19

In [9]:

# Preview CUSTOMER_Age_band
customer_age_band.preview(preview_table)

# Preview CUSTOMER_Age_band customer_age_band.preview(preview_table)

Out[9]:

	POINT_IN_TIME	GROCERYINVOICEITEMGUID	CUSTOMER_Age_band
0	2022-11-03 14:39:00	9faf5936-d4bb-4709-a530-c1624ec003a5	50-54
1	2023-02-15 21:10:22	5daf8edb-8625-4653-aaa0-9ac03df92017	70-74
2	2022-10-05 14:30:03	fc4769cf-41d6-4fb0-9cfc-fe30502cfa18	35-39
3	2022-11-09 18:07:20	37122b82-478b-4d9e-b236-69629b592c0b	25-29
4	2022-11-18 10:07:49	ae2ccf38-4e5d-4c76-b1e5-04f12307e45b	60-64
5	2023-03-19 15:22:18	dcdb7c98-dd62-4287-b432-bfe3a2317ebc	20-24
6	2022-08-16 15:37:21	6781acc6-652d-4867-a138-2d8adb278886	60-64
7	2023-04-19 15:25:05	8a60f455-aff1-4b4e-8c63-9ab89df2715d	70-74
8	2023-02-27 13:27:47	b807e05c-ff1c-4fb3-a760-e0e8ce29c859	35-39
9	2023-03-13 21:11:43	cfc620a2-0054-4b4b-99d5-040cf87cfe2d	15-19

In [10]:

# Save features to catalog 
customer_age.save()
customer_age_band.save()

# Save features to catalog customer_age.save() customer_age_band.save()

Done! |████████████████████████████████████████| 100% in 6.1s (0.16%/s)         
Done! |████████████████████████████████████████| 100% in 6.5s (0.16%/s)         

In [11]:

# Add description
customer_age.update_description("Age of the customer.")
customer_age_band.update_description("Age Band of the customer.")

# Add description customer_age.update_description("Age of the customer.") customer_age_band.update_description("Age Band of the customer.")

Check feature definition files (same as definition files we discussed in target tutorial)¶

In [12]:

customer_age.definition

customer_age.definition

Out[12]:

# Generated by SDK version: 1.0.2.dev46
from bson import ObjectId
from featurebyte import SCDTable
from featurebyte.api.request_column import RequestColumn


# scd_table name: "GROCERYCUSTOMER"
scd_table = SCDTable.get_by_id(ObjectId("662b5773aa13c89fa14554e0"))
scd_view = scd_table.get_view(
    view_mode="manual",
    drop_column_names=["record_available_at", "CurrentRecord"],
    column_cleaning_operations=[],
)
grouped = scd_view.as_features(
    column_names=["DateOfBirth"],
    feature_names=["CUSTOMER_DateOfBirth"],
    offset=None,
)
feat = grouped["CUSTOMER_DateOfBirth"]
request_col = RequestColumn.point_in_time()
feat_1 = ((request_col - feat).dt.day / 365.25).floor()
feat_1.name = "CUSTOMER_Age"
output = feat_1
output.save(_id=ObjectId("662b57fc1bdd8afa8a9be627"))

In [13]:

customer_age_band.definition

customer_age_band.definition

Out[13]:

# Generated by SDK version: 1.0.2.dev46
from bson import ObjectId
from featurebyte import SCDTable
from featurebyte.api.request_column import RequestColumn


# scd_table name: "GROCERYCUSTOMER"
scd_table = SCDTable.get_by_id(ObjectId("662b5773aa13c89fa14554e0"))
scd_view = scd_table.get_view(
    view_mode="manual",
    drop_column_names=["record_available_at", "CurrentRecord"],
    column_cleaning_operations=[],
)
grouped = scd_view.as_features(
    column_names=["DateOfBirth"],
    feature_names=["CUSTOMER_DateOfBirth"],
    offset=None,
)
feat = grouped["CUSTOMER_DateOfBirth"]
request_col = RequestColumn.point_in_time()
feat_1 = ((request_col - feat).dt.day / 365.25).floor()
feat_1.name = "CUSTOMER_Age"
feat_2 = ((((feat_1 + 1) / 5).ceil() - 1) * 5).astype(str) + "-"
feat_3 = ((((feat_1 + 1) / 5).ceil() - 1) * 5) + 4
feat_4 = feat_2 + feat_3.astype(str)
feat_4.name = "CUSTOMER_Age_band"
output = feat_4
output.save(_id=ObjectId("662b57fc1bdd8afa8a9be631"))

Concepts in this tutorial¶

• More on features
• More on lookup features
• More about point-in-time

SDK reference for¶

• Feature
• ViewColumn.as_feature()
• RequestColumn.point_in_time
• Feature.preview()
• Feature.save()

In [ ]: