Deploy and Serve a Feature List for Model Predictions¶

Once you have a feature list ready, it's essential to serve these features effectively to make real-time or batch predictions using your machine learning model.

In this section, we'll take the feature list we previously crafted and explore two primary ways to serve its values:

REST API: Ideal for real-time predictions where you need instantaneous results. For instance, in applications where user interactions require immediate feedback based on model predictions.

Batch Processing: Best suited for scenarios where you have a bulk of data and don't require instant results.

Important Note for FeatureByte Enterprise Users¶

In Catalogs with Approval Flow enabled, moving features to production-ready status involves a comprehensive approval process.

This includes several evaluations, such as checking the feature's compliance with default cleaning operations and the feature job setting of its source tables. It also involves confirming the status of these tables and backtesting the feature job setting to prevent future training-serving inconsistencies. Additionally, essential details of the feature, particularly its feature definition file, are shared and subjected to a thorough review.

In [1]:

import featurebyte as fb
from datetime import datetime

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

10:59:49 | WARNING  | Service endpoint is inaccessible: http://featurebyte-server:8088/
10:59:49 | INFO     | Using profile: tutorial
10:59:50 | INFO     | Using configuration file at: /Users/gxav/.featurebyte/config.yaml
10:59:50 | INFO     | Active profile: tutorial (https://tutorials.featurebyte.com/api/v1)
10:59:50 | INFO     | SDK version: 2.0.1.dev67
10:59:50 | INFO     | No catalog activated.
10:59:50 | INFO     | Catalog activated: Grocery Dataset Tutorial
16:13:11 | INFO     | Using configuration file at: /Users/gxav/.featurebyte/config.yaml
16:13:11 | INFO     | Active profile: tutorial (https://tutorials.featurebyte.com/api/v1)
16:13:11 | WARNING  | Remote SDK version (1.1.0.dev7) is different from local (1.1.0.dev1). Update local SDK to avoid unexpected behavior.
16:13:11 | INFO     | No catalog activated.
16:13:11 | 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	676239f4c3ba2239e045779a	Customer x ProductGroup Simple FeatureList	9	DRAFT	False	0.0	0.0	[GROCERYCUSTOMER, GROCERYINVOICE, INVOICEITEMS...	[customer, productgroup]	[customer, productgroup]	2024-12-18T02:57:16.277000

Get a feature list from Catalog¶

In [3]:

simple_feature_list = catalog.get_feature_list("Customer x ProductGroup Simple FeatureList")

simple_feature_list = catalog.get_feature_list("Customer x ProductGroup Simple FeatureList")

Loading Feature(s) |████████████████████████████████████████| 9/9 [100%] in 0.3s
Loading Feature(s) |████████████████████████████████████████| 9/9 [100%] in 0.2s

Deploy feature list¶

In [4]:

# Create a deployment
deployment = simple_feature_list.deploy(
  deployment_name="In-Store Customer x ProductGroup Spending 2w Spending - 9 features",
  make_production_ready=True,
)

# Create a deployment deployment = simple_feature_list.deploy( deployment_name="In-Store Customer x ProductGroup Spending 2w Spending - 9 features", make_production_ready=True, )

Done! |████████████████████████████████████████| 100% in 6.1s (0.17%/s)         
Done! |████████████████████████████████████████| 100% in 6.1s (0.17%/s)         
Done! |████████████████████████████████████████| 100% in 6.1s (0.17%/s)         
Done! |████████████████████████████████████████| 100% in 6.1s (0.17%/s)         

In [5]:

# Enable deployment
deployment.enable()

# Enable deployment deployment.enable()

Done! |████████████████████████████████████████| 100% in 2:53.2 (0.01%/s)       
Done! |████████████████████████████████████████| 100% in 1:49.2 (0.01%/s)       

In [6]:

# Check that the deployment is enabled
catalog.list_deployments()

# Check that the deployment is enabled catalog.list_deployments()

Out[6]:

	id	name	feature_list_name	feature_list_version	num_feature	enabled
0	67623aadf49d772007a631ad	In-Store Customer x ProductGroup Spending 2w S...	Customer x ProductGroup Simple FeatureList	V241218	9	True

In [7]:

# Check status of the feature list
print("simple_feature_list.status:", simple_feature_list.status)

# Check status of the feature list print("simple_feature_list.status:", simple_feature_list.status)

simple_feature_list.status: DEPLOYED

Get template for online serving¶

In [8]:

# Get a python template for consuming the feature serving API
deployment.get_online_serving_code(language="python")

# Get a python template for consuming the feature serving API deployment.get_online_serving_code(language="python")

Out[8]:

from typing import Any, Dict

import pandas as pd
import requests


def request_features(entity_serving_names: Dict[str, Any]) -> pd.DataFrame:
    """
    Send POST request to online serving endpoint

    Parameters
    ----------
    entity_serving_names: Dict[str, Any]
        Entity serving name values to used for serving request

    Returns
    -------
    pd.DataFrame
    """
    response = requests.post(
        url="https://tutorials.featurebyte.com/api/v1/deployment/67623aadf49d772007a631ad/online_features",
        headers={"Content-Type": "application/json", "active-catalog-id": "676225fc990c9e1a061d72ab", "Authorization": "Bearer alZJY8HvLxoi_emU_eukghMpSMzxL0eQ4ufUSWxvsLk"},
        json={"entity_serving_names": entity_serving_names},
    )
    assert response.status_code == 200, response.json()
    return pd.DataFrame.from_dict(response.json()["features"])


request_features([{"GROCERYCUSTOMERGUID": "016d7230-8602-42ef-9861-c672f48fb010", "PRODUCTGROUP": "Gla\u00e7ons"}])

In [9]:

# Get shell script
deployment.get_online_serving_code(language="sh")

# Get shell script deployment.get_online_serving_code(language="sh")

Out[9]:

#!/bin/sh

curl -X POST \
    -H 'Content-Type: application/json' \
    -H 'active-catalog-id: 676225fc990c9e1a061d72ab' \
    -H 'Authorization: Bearer alZJY8HvLxoi_emU_eukghMpSMzxL0eQ4ufUSWxvsLk' \
    -d '{"entity_serving_names": [{"GROCERYCUSTOMERGUID": "016d7230-8602-42ef-9861-c672f48fb010", "PRODUCTGROUP": "Gla\u00e7ons"}]}' \
    https://tutorials.featurebyte.com/api/v1/deployment/67623aadf49d772007a631ad/online_features

Create batch request table¶

Although batch request is not suitable for our "In-Store" use case, let's run a batch request for all current customers for a campaign for "Fromages".

In [10]:

# Get view of current customer
customer_table = catalog.get_table("GROCERYCUSTOMER")
customer_view = customer_table.get_view(view_mode="manual")
cond = customer_view.CurrentRecord == True
current_customer_view = customer_view[cond]

# Get view of current customer customer_table = catalog.get_table("GROCERYCUSTOMER") customer_view = customer_table.get_view(view_mode="manual") cond = customer_view.CurrentRecord == True current_customer_view = customer_view[cond]

In [11]:

# Check primary entity of the deployed feature list to obtain its serving name
simple_feature_list.primary_entity

# Check primary entity of the deployed feature list to obtain its serving name simple_feature_list.primary_entity

Out[11]:

[<featurebyte.api.entity.Entity at 0x329ecaed0>
 {
   'name': 'customer',
   'created_at': '2024-12-18T02:40:54.536000',
   'updated_at': '2024-12-18T02:40:58.932000',
   'description': None,
   'serving_names': [
     'GROCERYCUSTOMERGUID'
   ],
   'catalog_name': 'Grocery Dataset Tutorial'
 },
 <featurebyte.api.entity.Entity at 0x329ecaf70>
 {
   'name': 'productgroup',
   'created_at': '2024-12-18T02:40:55.766000',
   'updated_at': '2024-12-18T02:41:00.663000',
   'description': None,
   'serving_names': [
     'PRODUCTGROUP'
   ],
   'catalog_name': 'Grocery Dataset Tutorial'
 }]

In [12]:

# Add PRODUCTGROUP to the view with "Fromages" for a specific targeted campaign.
current_customer_view["PRODUCTGROUP"] = "Fromages"

# Add PRODUCTGROUP to the view with "Fromages" for a specific targeted campaign. current_customer_view["PRODUCTGROUP"] = "Fromages"

In [13]:

# Create batch request table from the view
batch_request_table = current_customer_view.create_batch_request_table(
    name="Current Customers for Fromages at " + datetime.now().strftime("%Y%m%d:%H%M"),
    columns=["GroceryCustomerGuid", "PRODUCTGROUP"],
    columns_rename_mapping={
        "GroceryCustomerGuid": "GROCERYCUSTOMERGUID",
    }
)

# Create batch request table from the view batch_request_table = current_customer_view.create_batch_request_table( name="Current Customers for Fromages at " + datetime.now().strftime("%Y%m%d:%H%M"), columns=["GroceryCustomerGuid", "PRODUCTGROUP"], columns_rename_mapping={ "GroceryCustomerGuid": "GROCERYCUSTOMERGUID", } )

Done! |████████████████████████████████████████| 100% in 9.1s (0.11%/s)         
Done! |████████████████████████████████████████| 100% in 12.2s (0.08%/s)        

In [14]:

# Get name of the batch request table
batch_request_table.name

# Get name of the batch request table batch_request_table.name

Out[14]:

'Current Customers for Fromages at 20241218:1103'

In [15]:

# List batch request tables in catalog
catalog.list_batch_request_tables()

# List batch request tables in catalog catalog.list_batch_request_tables()

Out[15]:

	id	name	type	shape	feature_store_name	created_at
0	67623b6bf49d772007a631ae	Current Customers for Fromages at 20241218:1103	view	[500, 2]	playground	2024-12-18T03:03:13.469000

Compute batch feature table¶

In [16]:

# Get deployment and batch request table
deployment = catalog.get_deployment("In-Store Customer x ProductGroup Spending 2w Spending - 9 features")
batch_request_table = catalog.get_batch_request_table(batch_request_table.name)

# Get deployment and batch request table deployment = catalog.get_deployment("In-Store Customer x ProductGroup Spending 2w Spending - 9 features") batch_request_table = catalog.get_batch_request_table(batch_request_table.name)

In [17]:

# Compute batch features
batch_features = deployment.compute_batch_feature_table(
    batch_request_table=batch_request_table,
    batch_feature_table_name = 
    f"Customer Simple FeatureList for Spending forecast with {batch_request_table.name}"
)

# Compute batch features batch_features = deployment.compute_batch_feature_table( batch_request_table=batch_request_table, batch_feature_table_name = f"Customer Simple FeatureList for Spending forecast with {batch_request_table.name}" )

Done! |████████████████████████████████████████| 100% in 9.1s (0.11%/s)         
Done! |████████████████████████████████████████| 100% in 12.1s (0.08%/s)        

In [18]:

# List observation tables
catalog.list_batch_feature_tables()

# List observation tables catalog.list_batch_feature_tables()

Out[18]:

	id	name	feature_store_name	batch_request_table_name	shape	created_at
0	67623b75f49d772007a631af	Customer Simple FeatureList for Spending forec...	playground	Current Customers for Fromages at 20241218:1103	[500, 11]	2024-12-18T03:03:21.865000

In [19]:

# Convert to pandas 
batch_features.to_pandas()

# Convert to pandas batch_features.to_pandas()

Downloading table |████████████████████████████████████████| 500/500 [100%] in 0

Out[19]:

	GROCERYCUSTOMERGUID	PRODUCTGROUP	CUSTOMER_Age_band	CUSTOMER_Latest_invoice_Amount	CUSTOMER_Count_of_invoice_14d	CUSTOMER_Avg_of_invoice_Amount_14d	CUSTOMER_Std_of_invoice_Amount_14d	CUSTOMER_Latest_invoice_Amount_Z_Score_to_invoice_Amount_28d	CUSTOMER_vs_OVERALL_item_TotalCost_across_product_ProductGroups_26w	CUSTOMER_x_PRODUCTGROUP_Sum_of_item_TotalCost_14d	CUSTOMER_x_PRODUCTGROUP_Time_Since_Latest_Timestamp
0	09f35825-38ef-4a01-8385-c41822f59de9	Fromages	50-54	41.78	NaN	NaN	NaN	1.000000	0.412312	NaN	1810.671115
1	7eee31c3-c152-4cc3-a87b-aeed2e015039	Fromages	70-74	8.00	1.0	8.000000	0.000000	-1.000000	0.343774	NaN	5365.382226
2	6a9598c3-26ba-4784-8551-9c20819c5690	Fromages	70-74	7.28	NaN	NaN	NaN	NaN	0.610161	NaN	1735.358338
3	8440debb-6abc-4adc-8c6c-749928141fd0	Fromages	20-24	106.18	1.0	106.180000	0.000000	1.000000	0.643433	7.5	15.810838
4	d3c61041-64b6-4f0b-8c89-c3aaf20aae61	Fromages	25-29	6.27	NaN	NaN	NaN	NaN	0.601962	NaN	2011.292782
...	...	...	...	...	...	...	...	...	...	...	...
495	bed95154-2bd8-4947-9090-54d14da66643	Fromages	35-39	54.05	NaN	NaN	NaN	NaN	0.522650	NaN	1048.096115
496	8c3066bd-f278-4f46-871b-7737f8847aaa	Fromages	40-44	7.58	NaN	NaN	NaN	-1.000000	0.822351	NaN	565.231949
497	69134758-d8c6-46e2-b2e5-6c8a3095d1d4	Fromages	30-34	9.79	1.0	9.790000	0.000000	NaN	0.467934	NaN	1189.893338
498	1b49a621-805a-4626-9f92-30077d7ef9bf	Fromages	65-69	6.49	2.0	4.840000	1.650000	-0.491692	0.351327	NaN	397.866393
499	3f922a2f-529e-4d8b-84a2-b9b5847ba455	Fromages	65-69	9.01	3.0	8.226667	5.202079	-0.344543	0.650721	NaN	2074.993893

500 rows × 11 columns

Downloading table |████████████████████████████████████████| 500/500 [100%] in 0

Out[19]:

	GROCERYCUSTOMERGUID	PRODUCTGROUP	CUSTOMER_Age_band	CUSTOMER_Latest_invoice_Amount	CUSTOMER_Count_of_invoice_14d	CUSTOMER_Avg_of_invoice_Amount_14d	CUSTOMER_Std_of_invoice_Amount_14d	CUSTOMER_Latest_invoice_Amount_Z_Score_to_invoice_Amount_28d	CUSTOMER_vs_OVERALL_item_TotalCost_across_product_ProductGroups_26w	CUSTOMER_x_PRODUCTGROUP_Sum_of_item_TotalCost_14d	CUSTOMER_x_PRODUCTGROUP_Time_Since_Latest_Timestamp
0	14f94723-4cae-4041-a11a-a044df1a5a5f	Fromages	60-64	14.28	NaN	NaN	NaN	1.000000	0.373275	NaN	1527.861052
1	d4559f7d-eb28-42c6-b47d-847de24952c2	Fromages	65-69	20.04	NaN	NaN	NaN	NaN	0.370545	NaN	2177.017718
2	1d363f8c-88f4-49eb-a107-4f1e2fbe6722	Fromages	25-29	73.16	NaN	NaN	NaN	NaN	0.781391	NaN	1118.766885
3	768e57df-7513-412b-a466-effd270ad6a6	Fromages	65-69	7.16	3.0	8.246667	4.395121	-0.644055	0.650126	NaN	1357.819107
4	48f6e550-ce63-457e-ad14-d511ee934661	Fromages	40-44	24.62	1.0	24.620000	0.000000	-0.219845	0.753619	NaN	475.681329
...	...	...	...	...	...	...	...	...	...	...	...
495	c1b3224a-5dbb-427d-87d3-3882c6bb1278	Fromages	50-54	6.69	NaN	NaN	NaN	NaN	0.104953	NaN	NaN
496	261d2979-0568-4544-8be0-32355049e565	Fromages	65-69	2.00	NaN	NaN	NaN	NaN	0.317982	NaN	NaN
497	825ee709-2f85-49e2-bec7-27bea06efe3a	Fromages	45-49	3.00	2.0	5.995000	2.995000	-0.990561	0.375129	NaN	NaN
498	ea68ce6b-b3c3-4ba5-a471-f11b1c5bd892	Fromages	45-49	4.50	NaN	NaN	NaN	NaN	NaN	NaN	NaN
499	b8b00613-2e48-4eed-8971-61718990a994	Fromages	45-49	10.00	2.0	6.000000	4.000000	1.264961	0.579119	NaN	NaN

500 rows × 11 columns

In [20]:

# download parquet file
batch_features.download()

# download parquet file batch_features.download()

Downloading table |████████████████████████████████████████| 500/500 [100%] in 0

Out[20]:

PosixPath('BATCH_FEATURE_TABLE_67623b7553faf734e68b1ed2.parquet')

Downloading table |████████████████████████████████████████| 500/500 [100%] in 0

Out[20]:

PosixPath('BATCH_FEATURE_TABLE_666959284c98806ac301afa0.parquet')

In [21]:

# delete if not needed any more
batch_features.delete()
batch_request_table.delete()

# delete if not needed any more batch_features.delete() batch_request_table.delete()

Done! |████████████████████████████████████████| 100% in 9.2s (0.11%/s)         
Done! |████████████████████████████████████████| 100% in 6.1s (0.17%/s)         
Done! |████████████████████████████████████████| 100% in 6.1s (0.17%/s)         
Done! |████████████████████████████████████████| 100% in 6.1s (0.17%/s)         

Manage Deployment¶

In [22]:

deployment = catalog.get_deployment("In-Store Customer x ProductGroup Spending 2w Spending - 9 features")

deployment = catalog.get_deployment("In-Store Customer x ProductGroup Spending 2w Spending - 9 features")

In [23]:

# get feature jobs status (this will produce meaningful results only once multiple jobs have been run)
# deployment.get_feature_jobs_status(job_history_window=12)

# get feature jobs status (this will produce meaningful results only once multiple jobs have been run) # deployment.get_feature_jobs_status(job_history_window=12)

Out[23]:

Job statistics (last 12 hours)

	request_date	job_history_window	job_duration_tolerance
0	2024-06-12T08:16:06.814724	12	60

	feature_name	aggregation_hash
0	CUSTOMER_Avg_of_invoice_Amount_14d	94bc6df7
1	CUSTOMER_Count_of_invoice_14d	2bc63818
2	CUSTOMER_Latest_invoice_Amount	d6976a85
3	CUSTOMER_Latest_invoice_Amount_Z_Score_to_invoice_Amount_28d	d6976a85
4	CUSTOMER_Latest_invoice_Amount_Z_Score_to_invoice_Amount_28d	94bc6df7
5	CUSTOMER_Latest_invoice_Amount_Z_Score_to_invoice_Amount_28d	906f9cb2
6	CUSTOMER_Std_of_invoice_Amount_14d	906f9cb2
7	CUSTOMER_vs_OVERALL_item_TotalCost_across_product_ProductGroups_26w	0b11a7c6
8	CUSTOMER_vs_OVERALL_item_TotalCost_across_product_ProductGroups_26w	76b8b9b9
9	CUSTOMER_x_PRODUCTGROUP_Sum_of_item_TotalCost_14d	1e1caeb1
10	CUSTOMER_x_PRODUCTGROUP_Time_Since_Latest_Timestamp	e7a12f5d

	aggregation_hash	frequency(min)	completed_jobs	max_duration(s)	95 percentile	frac_late	exceed_period	failed_jobs	incomplete_jobs	time_since_last
0	d6976a85	60	0	NaN	NaN	NaN	0	0	12	NaT
1	2bc63818	60	0	NaN	NaN	NaN	0	0	12	NaT
2	94bc6df7	60	0	NaN	NaN	NaN	0	0	12	NaT
3	906f9cb2	60	0	NaN	NaN	NaN	0	0	12	NaT
4	0b11a7c6	60	0	NaN	NaN	NaN	0	0	12	NaT
5	76b8b9b9	60	0	NaN	NaN	NaN	0	0	12	NaT
6	1e1caeb1	60	0	NaN	NaN	NaN	0	0	12	NaT
7	e7a12f5d	60	0	NaN	NaN	NaN	0	0	12	NaT

In [24]:

# Disable deployment
deployment.disable()

# Disable deployment deployment.disable()

Done! |████████████████████████████████████████| 100% in 15.2s (0.07%/s)        
Done! |████████████████████████████████████████| 100% in 12.1s (0.08%/s)        

In [25]:

# The deployment is still part of the catalog but disabled
catalog.list_deployments()

# The deployment is still part of the catalog but disabled catalog.list_deployments()

Out[25]:

	id	name	feature_list_name	feature_list_version	num_feature	enabled
0	67623aadf49d772007a631ad	In-Store Customer x ProductGroup Spending 2w S...	Customer x ProductGroup Simple FeatureList	V241218	9	False

Concepts in this tutorial¶

• Feature list deployment
• REST API and batch serving
• Feature jobs

SDK reference for¶

• Deployment
• Batch Request Table
• Batch Feature Table
• FeatureList.deploy()
• Deployment.enable()
• Deployment.get_online_serving_code()
• Deployment.compute batch feature_table()
• Table.get_view() in a manual mode
• View.create_batch_request_table()
• SourceTable.create_batch_request_table()