13. Compute Historical Feature Values

Compute historical feature values¶

Historical feature values are needed to train and test Machine Learning models.

Let's take the feature list we just created and compute feature values for a given observation table.

In [1]:

import featurebyte as fb

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

catalog_name = "Credit Default Dataset SDK Tutorial"
catalog = fb.Catalog.activate(catalog_name)

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

16:44:44 | WARNING  | Service endpoint is inaccessible: http://featurebyte-server:8088/
16:44:44 | INFO     | Using profile: tutorial
16:44:44 | INFO     | Using configuration file at: /Users/gxav/.featurebyte/config.yaml
16:44:44 | INFO     | Active profile: tutorial (https://tutorials.featurebyte.com/api/v1)
16:44:44 | INFO     | SDK version: 2.1.0.dev113
16:44:44 | INFO     | No catalog activated.
16:44:44 | INFO     | Catalog activated: Credit Default Dataset SDK Tutorial
16:12:25 | WARNING  | Remote SDK version (1.1.0.dev7) is different from local (1.1.0.dev1). Update local SDK to avoid unexpected behavior.
16:12:25 | INFO     | No catalog activated.
16:12:25 | INFO     | Catalog activated: Grocery Dataset Tutorial

List feature lists in Catalog¶

In [2]:

catalog.list_feature_lists()

catalog.list_feature_lists()

Out[2]:

	id	name	num_feature	status	deployed	readiness_frac	online_frac	tables	entities	primary_entity	created_at
0	67c2c8de53e241b5a68dd616	51 features for Credit Default	51	DRAFT	False	0.0	0.0	[NEW_APPLICATION, PRIOR_APPLICATIONS, CONSUMER...	[New Application, Client]	[New Application]	2025-03-01T08:44:22.006000

Get Feature List from Catalog¶

In [3]:

feature_list_name = "51 features for Credit Default"
simple_feature_list = catalog.get_feature_list(feature_list_name)

feature_list_name = "51 features for Credit Default" simple_feature_list = catalog.get_feature_list(feature_list_name)

Loading Feature(s) |████████████████████████████████████████| 51/51 [100%] in 0.

Get an observation table¶

In [4]:

# List observation tables
catalog.list_observation_tables()

# List observation tables catalog.list_observation_tables()

Out[4]:

	id	name	type	shape	feature_store_name	created_at
0	67c2c801fbb34f46dc0daa56	PREVIEW_TABLE	source_table	[50, 3]	playground	2025-03-01T08:40:43.630000
1	67c2c7f0fbb34f46dc0daa55	CREDIT_DEFAULT_EDA_2019_2023	source_table	[49963, 3]	playground	2025-03-01T08:40:27.363000
2	67c2c7e3fbb34f46dc0daa54	CREDIT_DEFAULT_HOLDOUT_2024_1H	source_table	[25666, 3]	playground	2025-03-01T08:40:12.763000
3	67c2c7cffbb34f46dc0daa53	CREDIT_DEFAULT_TRAIN_2019_2023	source_table	[256273, 3]	playground	2025-03-01T08:39:58.304000

In [5]:

# Get observation table: 'CREDIT_DEFAULT_TRAIN_2019_2023'
training_observations = catalog.get_observation_table("CREDIT_DEFAULT_TRAIN_2019_2023")

# Get observation table: 'CREDIT_DEFAULT_TRAIN_2019_2023' training_observations = catalog.get_observation_table("CREDIT_DEFAULT_TRAIN_2019_2023")

In [6]:

# Get observation table: 'CREDIT_DEFAULT_HOLDOUT_2024_1H'
holdout_observations = catalog.get_observation_table("CREDIT_DEFAULT_HOLDOUT_2024_1H")

# Get observation table: 'CREDIT_DEFAULT_HOLDOUT_2024_1H' holdout_observations = catalog.get_observation_table("CREDIT_DEFAULT_HOLDOUT_2024_1H")

Compute historical features¶

In [7]:

# Create training data
training_data_table = simple_feature_list.compute_historical_feature_table(
    training_observations,
    historical_feature_table_name=f"{feature_list_name} - TRAIN_2019_2023",
)

# Create training data training_data_table = simple_feature_list.compute_historical_feature_table( training_observations, historical_feature_table_name=f"{feature_list_name} - TRAIN_2019_2023", )

Done! |████████████████████████████████████████| 100% in 2:44.2 (0.01%/s)       
Done! |████████████████████████████████████████| 100% in 36.4s (0.03%/s)        

In [8]:

holdout_data_table = simple_feature_list.compute_historical_feature_table(
    holdout_observations,
    historical_feature_table_name=f"{feature_list_name} - HOLDOUT_2024_1H",
)

holdout_data_table = simple_feature_list.compute_historical_feature_table( holdout_observations, historical_feature_table_name=f"{feature_list_name} - HOLDOUT_2024_1H", )

Done! |████████████████████████████████████████| 100% in 1:55.4 (0.01%/s)       

In [9]:

### List historical feature tables from catalog
catalog.list_historical_feature_tables()

### List historical feature tables from catalog catalog.list_historical_feature_tables()

Out[9]:

	id	name	feature_store_name	observation_table_name	shape	created_at
0	67c2c9a34e08d83e21381725	51 features for Credit Default - HOLDOUT_2024_1H	playground	CREDIT_DEFAULT_HOLDOUT_2024_1H	[25666, 54]	2025-03-01T08:49:21.154000
1	67c2c8fe3df413286793fb5a	51 features for Credit Default - TRAIN_2019_2023	playground	CREDIT_DEFAULT_TRAIN_2019_2023	[256273, 54]	2025-03-01T08:47:25.134000

Concepts in this tutorial¶

• More on Historical feature serving

SDK reference for¶

• Historical feature table
• FeatureList.compute historical feature table()
• FeatureList.compute_historical_features() to compute directly a data frame