AdventureWorks - Predicting customer churn¶

In this notebook, we will demonstrate how getML can be used for a customer churn project using a synthetic dataset of a fictional company.

Summary:

Prediction type: Classification model
Domain: Customer loyalty
Prediction target: churn
Population size: 19704

Author: Dr. Patrick Urbanke

Background¶

AdventureWorks is a fictional company that sells bicycles. It is used by Microsoft to showcase how its MS SQL Server can be used to manage business data. Since this dataset resembles a real-world customer database and it is open-source, we will use it to showcase how getML can be used for a classic customer churn project (real customer databases are not easily available for the purposes of showcasing and benchmarking, for reasons of data privacy).

The dataset has been downloaded from the CTU Prague relational learning repository (Motl and Schulte, 2015).

Analysis¶

Let's get started with the analysis and set up your session:

In [1]:

  Copied!     
 
%pip install -q "getml==1.5.0" "pandas~=2.2" 

import numpy as np
import pandas as pd

import getml

getml.engine.launch()
getml.set_project('adventure_works')
%pip install -q "getml==1.5.0" "pandas~=2.2" import numpy as np import pandas as pd import getml getml.engine.launch() getml.set_project('adventure_works')

Note: you may need to restart the kernel to use updated packages.
Launching ./getML --allow-push-notifications=true --allow-remote-ips=false --home-directory=/home/user --in-memory=true --install=false --launch-browser=true --log=false in /home/user/.getML/getml-1.5.0-x64-community-edition-linux...
Launched the getML Engine. The log output will be stored in /home/user/.getML/logs/20240912122459.log.
  Loading pipelines... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:00

Connected to project 'adventure_works'.

1. Loading data¶

1.1 Download from source¶

We begin by downloading the data:

In [2]:

  Copied!     
 
conn = getml.database.connect_mariadb(
    host="db.relational-data.org",
    dbname="AdventureWorks2014",
    port=3306,
    user="guest",
    password="relational"
)

conn
conn = getml.database.connect_mariadb( host="db.relational-data.org", dbname="AdventureWorks2014", port=3306, user="guest", password="relational" ) conn

Out[2]:

Connection(dbname='AdventureWorks2014',
           dialect='mysql',
           host='db.relational-data.org',
           port=3306)

In [3]:

  Copied!     
 
def load_if_needed(name):
    """
    Loads the data from the relational learning
    repository, if the data frame has not already
    been loaded.
    """
    if not getml.data.exists(name):
        data_frame = getml.DataFrame.from_db(
            name=name,
            table_name=name,
            conn=conn
        )
        data_frame.save()
    else:
        data_frame = getml.data.load_data_frame(name)
    return data_frame
def load_if_needed(name): """ Loads the data from the relational learning repository, if the data frame has not already been loaded. """ if not getml.data.exists(name): data_frame = getml.DataFrame.from_db( name=name, table_name=name, conn=conn ) data_frame.save() else: data_frame = getml.data.load_data_frame(name) return data_frame

In [4]:

  Copied!     
 
product = load_if_needed("Product")
sales_order_detail = load_if_needed("SalesOrderDetail")
sales_order_header = load_if_needed("SalesOrderHeader")
sales_order_reason = load_if_needed("SalesOrderHeaderSalesReason")
special_offer = load_if_needed("SpecialOffer")
store = load_if_needed("Store")
product = load_if_needed("Product") sales_order_detail = load_if_needed("SalesOrderDetail") sales_order_header = load_if_needed("SalesOrderHeader") sales_order_reason = load_if_needed("SalesOrderHeaderSalesReason") special_offer = load_if_needed("SpecialOffer") store = load_if_needed("Store")

In [5]:

  Copied!     
 
product
product

Out[5]:

name	ProductID	MakeFlag	FinishedGoodsFlag	SafetyStockLevel	ReorderPoint	DaysToManufacture	ProductSubcategoryID	ProductModelID	Name	ProductNumber	Color	StandardCost	ListPrice	Size	SizeUnitMeasureCode	WeightUnitMeasureCode	Weight	ProductLine	Class	Style	SellStartDate	SellEndDate	DiscontinuedDate	rowguid	ModifiedDate
role	unused_float	unused_float	unused_float	unused_float	unused_float	unused_float	unused_float	unused_float	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string
0	1	0	0	1000	750	0	nan	nan	Adjustable Race	AR-5381	NULL	0.0000	0.0000	NULL	NULL	NULL	NULL	NULL	NULL	NULL	2008-04-30 00:00:00	NULL	NULL	694215B7-08F7-4C0D-ACB1-D734BA44...	2014-02-08 09:01:36
1	2	0	0	1000	750	0	nan	nan	Bearing Ball	BA-8327	NULL	0.0000	0.0000	NULL	NULL	NULL	NULL	NULL	NULL	NULL	2008-04-30 00:00:00	NULL	NULL	58AE3C20-4F3A-4749-A7D4-D568806C...	2014-02-08 09:01:36
2	3	1	0	800	600	1	nan	nan	BB Ball Bearing	BE-2349	NULL	0.0000	0.0000	NULL	NULL	NULL	NULL	NULL	NULL	NULL	2008-04-30 00:00:00	NULL	NULL	9C21AED2-5BFA-4F18-BCB8-F11638DC...	2014-02-08 09:01:36
3	4	0	0	800	600	0	nan	nan	Headset Ball Bearings	BE-2908	NULL	0.0000	0.0000	NULL	NULL	NULL	NULL	NULL	NULL	NULL	2008-04-30 00:00:00	NULL	NULL	ECFED6CB-51FF-49B5-B06C-7D8AC834...	2014-02-08 09:01:36
4	316	1	0	800	600	1	nan	nan	Blade	BL-2036	NULL	0.0000	0.0000	NULL	NULL	NULL	NULL	NULL	NULL	NULL	2008-04-30 00:00:00	NULL	NULL	E73E9750-603B-4131-89F5-3DD15ED5...	2014-02-08 09:01:36
	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
499	995	1	1	500	375	1	5	96	ML Bottom Bracket	BB-8107	NULL	44.9506	101.2400	NULL	NULL	G	168.00	NULL	M	NULL	2013-05-30 00:00:00	NULL	NULL	71AB847F-D091-42D6-B735-7B0C2D82...	2014-02-08 09:01:36
500	996	1	1	500	375	1	5	97	HL Bottom Bracket	BB-9108	NULL	53.9416	121.4900	NULL	NULL	G	170.00	NULL	H	NULL	2013-05-30 00:00:00	NULL	NULL	230C47C5-08B2-4CE3-B706-69C0BDD6...	2014-02-08 09:01:36
501	997	1	1	100	75	4	2	31	Road-750 Black, 44	BK-R19B-44	Black	343.6496	539.9900	44	CM	LB	19.77	R	L	U	2013-05-30 00:00:00	NULL	NULL	44CE4802-409F-43AB-9B27-CA534218...	2014-02-08 09:01:36
502	998	1	1	100	75	4	2	31	Road-750 Black, 48	BK-R19B-48	Black	343.6496	539.9900	48	CM	LB	20.13	R	L	U	2013-05-30 00:00:00	NULL	NULL	3DE9A212-1D49-40B6-B10A-F564D981...	2014-02-08 09:01:36
503	999	1	1	100	75	4	2	31	Road-750 Black, 52	BK-R19B-52	Black	343.6496	539.9900	52	CM	LB	20.42	R	L	U	2013-05-30 00:00:00	NULL	NULL	AE638923-2B67-4679-B90E-ABBAB17D...	2014-02-08 09:01:36

504 rows x 25 columns
memory usage: 0.17 MB
name: Product
type: getml.DataFrame

In [6]:

  Copied!     
 
sales_order_detail
sales_order_detail

Out[6]:

name	SalesOrderID	SalesOrderDetailID	OrderQty	ProductID	SpecialOfferID	CarrierTrackingNumber	UnitPrice	UnitPriceDiscount	LineTotal	rowguid	ModifiedDate
role	unused_float	unused_float	unused_float	unused_float	unused_float	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string
0	43659	1	1	776	1	4911-403C-98	2024.9940	0.0000	2024.994000	B207C96D-D9E6-402B-8470-2CC176C4...	2011-05-30 22:00:00
1	43659	2	3	777	1	4911-403C-98	2024.9940	0.0000	6074.982000	7ABB600D-1E77-41BE-9FE5-B9142CFC...	2011-05-30 22:00:00
2	43659	3	1	778	1	4911-403C-98	2024.9940	0.0000	2024.994000	475CF8C6-49F6-486E-B0AD-AFC6A50C...	2011-05-30 22:00:00
3	43659	4	1	771	1	4911-403C-98	2039.9940	0.0000	2039.994000	04C4DE91-5815-45D6-8670-F462719F...	2011-05-30 22:00:00
4	43659	5	1	772	1	4911-403C-98	2039.9940	0.0000	2039.994000	5A74C7D2-E641-438E-A7AC-37BF2328...	2011-05-30 22:00:00
	...	...	...	...	...	...	...	...	...	...	...
121312	75122	121313	1	878	1	NULL	21.9800	0.0000	21.980000	8CAD6675-18CC-4F47-8287-97B41A8E...	2014-06-29 22:00:00
121313	75122	121314	1	712	1	NULL	8.9900	0.0000	8.990000	84F1C363-1C50-4442-BE16-541C59B6...	2014-06-29 22:00:00
121314	75123	121315	1	878	1	NULL	21.9800	0.0000	21.980000	C18B6476-429F-4BB1-828E-2BE5F82A...	2014-06-29 22:00:00
121315	75123	121316	1	879	1	NULL	159.0000	0.0000	159.000000	75A89C6A-C60A-47EA-8A52-B52A9C43...	2014-06-29 22:00:00
121316	75123	121317	1	712	1	NULL	8.9900	0.0000	8.990000	73646D26-0461-450D-8019-2C6C8586...	2014-06-29 22:00:00

121317 rows x 11 columns
memory usage: 21.60 MB
name: SalesOrderDetail
type: getml.DataFrame

In [7]:

  Copied!     
 
sales_order_header
sales_order_header

Out[7]:

name	SalesOrderID	RevisionNumber	Status	OnlineOrderFlag	CustomerID	SalesPersonID	TerritoryID	BillToAddressID	ShipToAddressID	ShipMethodID	CreditCardID	CurrencyRateID	OrderDate	DueDate	ShipDate	SalesOrderNumber	PurchaseOrderNumber	AccountNumber	CreditCardApprovalCode	SubTotal	TaxAmt	Freight	TotalDue	Comment	rowguid	ModifiedDate
role	unused_float	unused_float	unused_float	unused_float	unused_float	unused_float	unused_float	unused_float	unused_float	unused_float	unused_float	unused_float	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string
0	43659	8	5	0	29825	279	5	985	985	5	16281	nan	2011-05-30 22:00:00	2011-06-12 00:00:00	2011-06-07 00:00:00	SO43659	PO522145787	10-4020-000676	105041Vi84182	20565.6206	1971.5149	616.0984	23153.2339	NULL	79B65321-39CA-4115-9CBA-8FE0903E...	2011-06-06 22:00:00
1	43660	8	5	0	29672	279	5	921	921	5	5618	nan	2011-05-30 22:00:00	2011-06-12 00:00:00	2011-06-07 00:00:00	SO43660	PO18850127500	10-4020-000117	115213Vi29411	1294.2529	124.2483	38.8276	1457.3288	NULL	738DC42D-D03B-48A1-9822-F95A67EA...	2011-06-06 22:00:00
2	43661	8	5	0	29734	282	6	517	517	5	1346	4	2011-05-30 22:00:00	2011-06-12 00:00:00	2011-06-07 00:00:00	SO43661	PO18473189620	10-4020-000442	85274Vi6854	32726.4786	3153.7696	985.5530	36865.8012	NULL	D91B9131-18A4-4A11-BC3A-90B6F53E...	2011-06-06 22:00:00
3	43662	8	5	0	29994	282	6	482	482	5	10456	4	2011-05-30 22:00:00	2011-06-12 00:00:00	2011-06-07 00:00:00	SO43662	PO18444174044	10-4020-000227	125295Vi53935	28832.5289	2775.1646	867.2389	32474.9324	NULL	4A1ECFC0-CC3A-4740-B028-1C50BB48...	2011-06-06 22:00:00
4	43663	8	5	0	29565	276	4	1073	1073	5	4322	nan	2011-05-30 22:00:00	2011-06-12 00:00:00	2011-06-07 00:00:00	SO43663	PO18009186470	10-4020-000510	45303Vi22691	419.4589	40.2681	12.5838	472.3108	NULL	9B1E7A40-6AE0-4AD3-811C-A6495185...	2011-06-06 22:00:00
	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
31460	75119	8	5	1	11981	nan	1	17649	17649	1	6761	nan	2014-06-29 22:00:00	2014-07-12 00:00:00	2014-07-07 00:00:00	SO75119	NULL	10-4030-011981	429826Vi35166	42.2800	3.3824	1.0570	46.7194	NULL	9382F1C9-383A-435F-9449-0EECEA21...	2014-07-06 22:00:00
31461	75120	8	5	1	18749	nan	6	28374	28374	1	8925	nan	2014-06-29 22:00:00	2014-07-12 00:00:00	2014-07-07 00:00:00	SO75120	NULL	10-4030-018749	929849Vi46003	84.9600	6.7968	2.1240	93.8808	NULL	AE6A4FCF-FF73-4CD4-AF2C-5993D00D...	2014-07-06 22:00:00
31462	75121	8	5	1	15251	nan	6	26553	26553	1	14220	nan	2014-06-29 22:00:00	2014-07-12 00:00:00	2014-07-07 00:00:00	SO75121	NULL	10-4030-015251	529864Vi73738	74.9800	5.9984	1.8745	82.8529	NULL	D7395C0E-00CB-4BFA-A238-0D6A9F49...	2014-07-06 22:00:00
31463	75122	8	5	1	15868	nan	6	14616	14616	1	18719	nan	2014-06-29 22:00:00	2014-07-12 00:00:00	2014-07-07 00:00:00	SO75122	NULL	10-4030-015868	330022Vi97312	30.9700	2.4776	0.7743	34.2219	NULL	4221035A-4159-492F-AF40-4363A64F...	2014-07-06 22:00:00
31464	75123	8	5	1	18759	nan	6	14024	14024	1	10084	nan	2014-06-29 22:00:00	2014-07-12 00:00:00	2014-07-07 00:00:00	SO75123	NULL	10-4030-018759	230370Vi51970	189.9700	15.1976	4.7493	209.9169	NULL	D54752FF-2B54-4BE5-95EA-3B72289C...	2014-07-06 22:00:00

31465 rows x 26 columns
memory usage: 12.56 MB
name: SalesOrderHeader
type: getml.DataFrame

In [8]:

  Copied!     
 
sales_order_reason
sales_order_reason

Out[8]:

name	SalesOrderID	SalesReasonID	ModifiedDate
role	unused_float	unused_float	unused_string
0	43697	5	2011-05-30 22:00:00
1	43697	9	2011-05-30 22:00:00
2	43702	5	2011-05-31 22:00:00
3	43702	9	2011-05-31 22:00:00
4	43703	5	2011-05-31 22:00:00
	...	...	...
27642	75119	1	2014-06-29 22:00:00
27643	75120	1	2014-06-29 22:00:00
27644	75121	1	2014-06-29 22:00:00
27645	75122	1	2014-06-29 22:00:00
27646	75123	1	2014-06-29 22:00:00

27647 rows x 3 columns
memory usage: 1.22 MB
name: SalesOrderHeaderSalesReason
type: getml.DataFrame

In [9]:

  Copied!     
 
special_offer
special_offer

Out[9]:

name	SpecialOfferID	MinQty	MaxQty	Description	DiscountPct	Type	Category	StartDate	EndDate	rowguid	ModifiedDate
role	unused_float	unused_float	unused_float	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string
0	1	0	nan	No Discount	0.0000	No Discount	No Discount	2011-05-01 00:00:00	2014-11-30 00:00:00	0290C4F5-191F-4337-AB6B-0A2DDE03...	2011-03-31 22:00:00
1	2	11	14	Volume Discount 11 to 14	0.0200	Volume Discount	Reseller	2011-05-31 00:00:00	2014-05-30 00:00:00	D7542EE7-15DB-4541-985C-5CC27AEF...	2011-04-30 22:00:00
2	3	15	24	Volume Discount 15 to 24	0.0500	Volume Discount	Reseller	2011-05-31 00:00:00	2014-05-30 00:00:00	4BDBCC01-8CF7-40A9-B643-40EC5B71...	2011-04-30 22:00:00
3	4	25	40	Volume Discount 25 to 40	0.1000	Volume Discount	Reseller	2011-05-31 00:00:00	2014-05-30 00:00:00	504B5E85-8F3F-4EBC-9E1D-C1BC5DEA...	2011-04-30 22:00:00
4	5	41	60	Volume Discount 41 to 60	0.1500	Volume Discount	Reseller	2011-05-31 00:00:00	2014-05-30 00:00:00	677E1D9D-944F-4E81-90E8-47EB0A82...	2011-04-30 22:00:00
	...	...	...	...	...	...	...	...	...	...	...
11	12	0	nan	LL Road Frame Sale	0.3500	Excess Inventory	Reseller	2013-05-30 00:00:00	2013-07-14 00:00:00	C0AF1C89-9722-4235-9248-3FBA4D9E...	2013-04-29 22:00:00
12	13	0	nan	Touring-3000 Promotion	0.1500	New Product	Reseller	2013-05-30 00:00:00	2013-08-29 00:00:00	5061CCE4-E021-45A8-9A75-DFB36CBB...	2013-04-29 22:00:00
13	14	0	nan	Touring-1000 Promotion	0.2000	New Product	Reseller	2013-05-30 00:00:00	2013-08-29 00:00:00	1AF84A9E-A98C-4BD9-B48F-DC2B8B6B...	2013-04-29 22:00:00
14	15	0	nan	Half-Price Pedal Sale	0.5000	Seasonal Discount	Customer	2013-07-14 00:00:00	2013-08-14 00:00:00	03E3594D-6EBB-46A6-B8EE-A9289C0C...	2013-06-13 22:00:00
15	16	0	nan	Mountain-500 Silver Clearance Sa...	0.4000	Discontinued Product	Reseller	2014-03-31 00:00:00	2014-05-30 00:00:00	EB7CB484-BCCF-4D2D-BF73-521B2001...	2014-02-28 23:00:00

16 rows x 11 columns
memory usage: 0.00 MB
name: SpecialOffer
type: getml.DataFrame

In [10]:

  Copied!     
 
store
store

Out[10]:

name	BusinessEntityID	SalesPersonID	Name	Demographics	rowguid	ModifiedDate
role	unused_float	unused_float	unused_string	unused_string	unused_string	unused_string
0	292	279	Next-Door Bike Store	<StoreSurvey xmlns="http://schem...	A22517E3-848D-4EBE-B9D9-7437F343...	2014-09-12 09:15:07
1	294	276	Professional Sales and Service	<StoreSurvey xmlns="http://schem...	B50CA50B-C601-4A13-B07E-2C63862D...	2014-09-12 09:15:07
2	296	277	Riders Company	<StoreSurvey xmlns="http://schem...	337C3688-1339-4E1A-A08A-B54B2356...	2014-09-12 09:15:07
3	298	275	The Bike Mechanics	<StoreSurvey xmlns="http://schem...	7894F278-F0C8-4D16-BD75-213FDBF1...	2014-09-12 09:15:07
4	300	286	Nationwide Supply	<StoreSurvey xmlns="http://schem...	C3FC9705-A8C4-4F3A-9550-EB2FA4B7...	2014-09-12 09:15:07
	...	...	...	...	...	...
696	1988	282	Retreat Inn	<StoreSurvey xmlns="http://schem...	EA21EC81-1BFA-4A07-9B4D-73D9852A...	2014-09-12 09:15:07
697	1990	281	Technical Parts Manufacturing	<StoreSurvey xmlns="http://schem...	C8E3C4ED-8F58-4DB2-B600-E0CD11D9...	2014-09-12 09:15:07
698	1992	277	Totes & Baskets Company	<StoreSurvey xmlns="http://schem...	CE860B58-643C-4567-BFD8-06E97969...	2014-09-12 09:15:07
699	1994	277	World of Bikes	<StoreSurvey xmlns="http://schem...	0C10F2B6-A13A-440C-9C25-5B28D482...	2014-09-12 09:15:07
700	2051	275	A Bicycle Association	<StoreSurvey xmlns="http://schem...	82237172-D3FE-4A95-82EF-636F6552...	2014-09-12 09:15:07

701 rows x 6 columns
memory usage: 0.38 MB
name: Store
type: getml.DataFrame

1.2 Prepare data for getML¶

getML requires that we define roles for each of the columns.

In [11]:

  Copied!     
 
product.set_role("ProductID", getml.data.roles.join_key)
product.set_role(["MakeFlag", "ProductSubcategoryID", "ProductModelID"], getml.data.roles.categorical)
product.set_role(["SafetyStockLevel", "ReorderPoint", "StandardCost", "ListPrice"], getml.data.roles.numerical)

product
product.set_role("ProductID", getml.data.roles.join_key) product.set_role(["MakeFlag", "ProductSubcategoryID", "ProductModelID"], getml.data.roles.categorical) product.set_role(["SafetyStockLevel", "ReorderPoint", "StandardCost", "ListPrice"], getml.data.roles.numerical) product

Out[11]:

name	ProductID	MakeFlag	ProductSubcategoryID	ProductModelID	SafetyStockLevel	ReorderPoint	StandardCost	ListPrice	FinishedGoodsFlag	DaysToManufacture	Name	ProductNumber	Color	Size	SizeUnitMeasureCode	WeightUnitMeasureCode	Weight	ProductLine	Class	Style	SellStartDate	SellEndDate	DiscontinuedDate	rowguid	ModifiedDate
role	join_key	categorical	categorical	categorical	numerical	numerical	numerical	numerical	unused_float	unused_float	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string
0	1	0	nan	nan	1000	750	0	0	0	0	Adjustable Race	AR-5381	NULL	NULL	NULL	NULL	NULL	NULL	NULL	NULL	2008-04-30 00:00:00	NULL	NULL	694215B7-08F7-4C0D-ACB1-D734BA44...	2014-02-08 09:01:36
1	2	0	nan	nan	1000	750	0	0	0	0	Bearing Ball	BA-8327	NULL	NULL	NULL	NULL	NULL	NULL	NULL	NULL	2008-04-30 00:00:00	NULL	NULL	58AE3C20-4F3A-4749-A7D4-D568806C...	2014-02-08 09:01:36
2	3	1	nan	nan	800	600	0	0	0	1	BB Ball Bearing	BE-2349	NULL	NULL	NULL	NULL	NULL	NULL	NULL	NULL	2008-04-30 00:00:00	NULL	NULL	9C21AED2-5BFA-4F18-BCB8-F11638DC...	2014-02-08 09:01:36
3	4	0	nan	nan	800	600	0	0	0	0	Headset Ball Bearings	BE-2908	NULL	NULL	NULL	NULL	NULL	NULL	NULL	NULL	2008-04-30 00:00:00	NULL	NULL	ECFED6CB-51FF-49B5-B06C-7D8AC834...	2014-02-08 09:01:36
4	316	1	nan	nan	800	600	0	0	0	1	Blade	BL-2036	NULL	NULL	NULL	NULL	NULL	NULL	NULL	NULL	2008-04-30 00:00:00	NULL	NULL	E73E9750-603B-4131-89F5-3DD15ED5...	2014-02-08 09:01:36
	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
499	995	1	5	96	500	375	44.9506	101.24	1	1	ML Bottom Bracket	BB-8107	NULL	NULL	NULL	G	168.00	NULL	M	NULL	2013-05-30 00:00:00	NULL	NULL	71AB847F-D091-42D6-B735-7B0C2D82...	2014-02-08 09:01:36
500	996	1	5	97	500	375	53.9416	121.49	1	1	HL Bottom Bracket	BB-9108	NULL	NULL	NULL	G	170.00	NULL	H	NULL	2013-05-30 00:00:00	NULL	NULL	230C47C5-08B2-4CE3-B706-69C0BDD6...	2014-02-08 09:01:36
501	997	1	2	31	100	75	343.6496	539.99	1	4	Road-750 Black, 44	BK-R19B-44	Black	44	CM	LB	19.77	R	L	U	2013-05-30 00:00:00	NULL	NULL	44CE4802-409F-43AB-9B27-CA534218...	2014-02-08 09:01:36
502	998	1	2	31	100	75	343.6496	539.99	1	4	Road-750 Black, 48	BK-R19B-48	Black	48	CM	LB	20.13	R	L	U	2013-05-30 00:00:00	NULL	NULL	3DE9A212-1D49-40B6-B10A-F564D981...	2014-02-08 09:01:36
503	999	1	2	31	100	75	343.6496	539.99	1	4	Road-750 Black, 52	BK-R19B-52	Black	52	CM	LB	20.42	R	L	U	2013-05-30 00:00:00	NULL	NULL	AE638923-2B67-4679-B90E-ABBAB17D...	2014-02-08 09:01:36

504 rows x 25 columns
memory usage: 0.16 MB
name: Product
type: getml.DataFrame

In [12]:

  Copied!     
 
sales_order_detail.set_role(["SalesOrderID", "SalesOrderDetailID", "ProductID", "SpecialOfferID"], getml.data.roles.join_key)
sales_order_detail.set_role(["OrderQty", "UnitPrice", "UnitPriceDiscount", "LineTotal"], getml.data.roles.numerical)
sales_order_detail.set_role("ModifiedDate", getml.data.roles.time_stamp)

sales_order_detail
sales_order_detail.set_role(["SalesOrderID", "SalesOrderDetailID", "ProductID", "SpecialOfferID"], getml.data.roles.join_key) sales_order_detail.set_role(["OrderQty", "UnitPrice", "UnitPriceDiscount", "LineTotal"], getml.data.roles.numerical) sales_order_detail.set_role("ModifiedDate", getml.data.roles.time_stamp) sales_order_detail

Out[12]:

name	ModifiedDate	SalesOrderID	SalesOrderDetailID	ProductID	SpecialOfferID	OrderQty	UnitPrice	UnitPriceDiscount	LineTotal	CarrierTrackingNumber	rowguid
role	time_stamp	join_key	join_key	join_key	join_key	numerical	numerical	numerical	numerical	unused_string	unused_string
unit	time stamp, comparison only
0	2011-05-30 22:00:00	43659	1	776	1	1	2024.994	0	2024.994	4911-403C-98	B207C96D-D9E6-402B-8470-2CC176C4...
1	2011-05-30 22:00:00	43659	2	777	1	3	2024.994	0	6074.982	4911-403C-98	7ABB600D-1E77-41BE-9FE5-B9142CFC...
2	2011-05-30 22:00:00	43659	3	778	1	1	2024.994	0	2024.994	4911-403C-98	475CF8C6-49F6-486E-B0AD-AFC6A50C...
3	2011-05-30 22:00:00	43659	4	771	1	1	2039.994	0	2039.994	4911-403C-98	04C4DE91-5815-45D6-8670-F462719F...
4	2011-05-30 22:00:00	43659	5	772	1	1	2039.994	0	2039.994	4911-403C-98	5A74C7D2-E641-438E-A7AC-37BF2328...
	...	...	...	...	...	...	...	...	...	...	...
121312	2014-06-29 22:00:00	75122	121313	878	1	1	21.98	0	21.98	NULL	8CAD6675-18CC-4F47-8287-97B41A8E...
121313	2014-06-29 22:00:00	75122	121314	712	1	1	8.99	0	8.99	NULL	84F1C363-1C50-4442-BE16-541C59B6...
121314	2014-06-29 22:00:00	75123	121315	878	1	1	21.98	0	21.98	NULL	C18B6476-429F-4BB1-828E-2BE5F82A...
121315	2014-06-29 22:00:00	75123	121316	879	1	1	159	0	159	NULL	75A89C6A-C60A-47EA-8A52-B52A9C43...
121316	2014-06-29 22:00:00	75123	121317	712	1	1	8.99	0	8.99	NULL	73646D26-0461-450D-8019-2C6C8586...

121317 rows x 11 columns
memory usage: 14.08 MB
name: SalesOrderDetail
type: getml.DataFrame

In [13]:

  Copied!     
 
sales_order_reason.set_role("SalesOrderID", getml.data.roles.join_key)
sales_order_reason.set_role("SalesReasonID", getml.data.roles.categorical)

sales_order_reason
sales_order_reason.set_role("SalesOrderID", getml.data.roles.join_key) sales_order_reason.set_role("SalesReasonID", getml.data.roles.categorical) sales_order_reason

Out[13]:

name	SalesOrderID	SalesReasonID	ModifiedDate
role	join_key	categorical	unused_string
0	43697	5	2011-05-30 22:00:00
1	43697	9	2011-05-30 22:00:00
2	43702	5	2011-05-31 22:00:00
3	43702	9	2011-05-31 22:00:00
4	43703	5	2011-05-31 22:00:00
	...	...	...
27642	75119	1	2014-06-29 22:00:00
27643	75120	1	2014-06-29 22:00:00
27644	75121	1	2014-06-29 22:00:00
27645	75122	1	2014-06-29 22:00:00
27646	75123	1	2014-06-29 22:00:00

27647 rows x 3 columns
memory usage: 1.00 MB
name: SalesOrderHeaderSalesReason
type: getml.DataFrame

In [14]:

  Copied!     
 
special_offer.set_role(["SpecialOfferID"], getml.data.roles.join_key)
special_offer.set_role(["MinQty", "DiscountPct"], getml.data.roles.numerical)
special_offer.set_role(["Category", "Description", "Type"], getml.data.roles.categorical)
special_offer.set_role(["StartDate", "EndDate"], getml.data.roles.time_stamp)

special_offer
special_offer.set_role(["SpecialOfferID"], getml.data.roles.join_key) special_offer.set_role(["MinQty", "DiscountPct"], getml.data.roles.numerical) special_offer.set_role(["Category", "Description", "Type"], getml.data.roles.categorical) special_offer.set_role(["StartDate", "EndDate"], getml.data.roles.time_stamp) special_offer

Out[14]:

name	StartDate	EndDate	SpecialOfferID	Category	Description	Type	MinQty	DiscountPct	MaxQty	rowguid	ModifiedDate
role	time_stamp	time_stamp	join_key	categorical	categorical	categorical	numerical	numerical	unused_float	unused_string	unused_string
unit	time stamp, comparison only	time stamp, comparison only
0	2011-05-01	2014-11-30	1	No Discount	No Discount	No Discount	0	0	nan	0290C4F5-191F-4337-AB6B-0A2DDE03...	2011-03-31 22:00:00
1	2011-05-31	2014-05-30	2	Reseller	Volume Discount 11 to 14	Volume Discount	11	0.02	14	D7542EE7-15DB-4541-985C-5CC27AEF...	2011-04-30 22:00:00
2	2011-05-31	2014-05-30	3	Reseller	Volume Discount 15 to 24	Volume Discount	15	0.05	24	4BDBCC01-8CF7-40A9-B643-40EC5B71...	2011-04-30 22:00:00
3	2011-05-31	2014-05-30	4	Reseller	Volume Discount 25 to 40	Volume Discount	25	0.1	40	504B5E85-8F3F-4EBC-9E1D-C1BC5DEA...	2011-04-30 22:00:00
4	2011-05-31	2014-05-30	5	Reseller	Volume Discount 41 to 60	Volume Discount	41	0.15	60	677E1D9D-944F-4E81-90E8-47EB0A82...	2011-04-30 22:00:00
	...	...	...	...	...	...	...	...	...	...	...
11	2013-05-30	2013-07-14	12	Reseller	LL Road Frame Sale	Excess Inventory	0	0.35	nan	C0AF1C89-9722-4235-9248-3FBA4D9E...	2013-04-29 22:00:00
12	2013-05-30	2013-08-29	13	Reseller	Touring-3000 Promotion	New Product	0	0.15	nan	5061CCE4-E021-45A8-9A75-DFB36CBB...	2013-04-29 22:00:00
13	2013-05-30	2013-08-29	14	Reseller	Touring-1000 Promotion	New Product	0	0.2	nan	1AF84A9E-A98C-4BD9-B48F-DC2B8B6B...	2013-04-29 22:00:00
14	2013-07-14	2013-08-14	15	Customer	Half-Price Pedal Sale	Seasonal Discount	0	0.5	nan	03E3594D-6EBB-46A6-B8EE-A9289C0C...	2013-06-13 22:00:00
15	2014-03-31	2014-05-30	16	Reseller	Mountain-500 Silver Clearance Sa...	Discontinued Product	0	0.4	nan	EB7CB484-BCCF-4D2D-BF73-521B2001...	2014-02-28 23:00:00

16 rows x 11 columns
memory usage: 0.00 MB
name: SpecialOffer
type: getml.DataFrame

In [15]:

  Copied!     
 
store.set_role(["SalesPersonID"], getml.data.roles.join_key)

store["test"] = store["ModifiedDate"].update(getml.data.random() > 0.5, "NULL")

store.set_role(["SalesPersonID"], getml.data.roles.join_key)
store.set_role(["test"], getml.data.roles.time_stamp)

store
store.set_role(["SalesPersonID"], getml.data.roles.join_key) store["test"] = store["ModifiedDate"].update(getml.data.random() > 0.5, "NULL") store.set_role(["SalesPersonID"], getml.data.roles.join_key) store.set_role(["test"], getml.data.roles.time_stamp) store

Out[15]:

name	test	SalesPersonID	BusinessEntityID	Name	Demographics	rowguid	ModifiedDate
role	time_stamp	join_key	unused_float	unused_string	unused_string	unused_string	unused_string
unit	time stamp, comparison only
0	2014-09-12 09:15:07	279	292	Next-Door Bike Store	<StoreSurvey xmlns="http://schem...	A22517E3-848D-4EBE-B9D9-7437F343...	2014-09-12 09:15:07
1	NULL	276	294	Professional Sales and Service	<StoreSurvey xmlns="http://schem...	B50CA50B-C601-4A13-B07E-2C63862D...	2014-09-12 09:15:07
2	2014-09-12 09:15:07	277	296	Riders Company	<StoreSurvey xmlns="http://schem...	337C3688-1339-4E1A-A08A-B54B2356...	2014-09-12 09:15:07
3	NULL	275	298	The Bike Mechanics	<StoreSurvey xmlns="http://schem...	7894F278-F0C8-4D16-BD75-213FDBF1...	2014-09-12 09:15:07
4	NULL	286	300	Nationwide Supply	<StoreSurvey xmlns="http://schem...	C3FC9705-A8C4-4F3A-9550-EB2FA4B7...	2014-09-12 09:15:07
	...	...	...	...	...	...	...
696	NULL	282	1988	Retreat Inn	<StoreSurvey xmlns="http://schem...	EA21EC81-1BFA-4A07-9B4D-73D9852A...	2014-09-12 09:15:07
697	2014-09-12 09:15:07	281	1990	Technical Parts Manufacturing	<StoreSurvey xmlns="http://schem...	C8E3C4ED-8F58-4DB2-B600-E0CD11D9...	2014-09-12 09:15:07
698	NULL	277	1992	Totes & Baskets Company	<StoreSurvey xmlns="http://schem...	CE860B58-643C-4567-BFD8-06E97969...	2014-09-12 09:15:07
699	NULL	277	1994	World of Bikes	<StoreSurvey xmlns="http://schem...	0C10F2B6-A13A-440C-9C25-5B28D482...	2014-09-12 09:15:07
700	2014-09-12 09:15:07	275	2051	A Bicycle Association	<StoreSurvey xmlns="http://schem...	82237172-D3FE-4A95-82EF-636F6552...	2014-09-12 09:15:07

701 rows x 7 columns
memory usage: 0.38 MB
name: Store
type: getml.DataFrame

In [16]:

  Copied!     
 
sales_order_header["SalesPersonIDCat"] = sales_order_header["SalesPersonID"]
sales_order_header["TerritoryIDCat"] = sales_order_header["TerritoryID"]

sales_order_header.set_role(["CustomerID", "SalesOrderID", "SalesPersonID", "TerritoryID"], getml.data.roles.join_key)
sales_order_header.set_role(
    ["RevisionNumber", "OnlineOrderFlag", "SalesPersonIDCat", "TerritoryIDCat", "ShipMethodID"], 
    getml.data.roles.categorical)
sales_order_header.set_role(["SubTotal", "TaxAmt", "Freight", "TotalDue"], getml.data.roles.numerical)
sales_order_header.set_role(["OrderDate", "DueDate", "ShipDate", "ModifiedDate"], getml.data.roles.time_stamp)

sales_order_header
sales_order_header["SalesPersonIDCat"] = sales_order_header["SalesPersonID"] sales_order_header["TerritoryIDCat"] = sales_order_header["TerritoryID"] sales_order_header.set_role(["CustomerID", "SalesOrderID", "SalesPersonID", "TerritoryID"], getml.data.roles.join_key) sales_order_header.set_role( ["RevisionNumber", "OnlineOrderFlag", "SalesPersonIDCat", "TerritoryIDCat", "ShipMethodID"], getml.data.roles.categorical) sales_order_header.set_role(["SubTotal", "TaxAmt", "Freight", "TotalDue"], getml.data.roles.numerical) sales_order_header.set_role(["OrderDate", "DueDate", "ShipDate", "ModifiedDate"], getml.data.roles.time_stamp) sales_order_header

Out[16]:

name	OrderDate	DueDate	ShipDate	ModifiedDate	CustomerID	SalesOrderID	SalesPersonID	TerritoryID	RevisionNumber	OnlineOrderFlag	SalesPersonIDCat	TerritoryIDCat	ShipMethodID	SubTotal	TaxAmt	Freight	TotalDue	Status	BillToAddressID	ShipToAddressID	CreditCardID	CurrencyRateID	SalesOrderNumber	PurchaseOrderNumber	AccountNumber	CreditCardApprovalCode	Comment	rowguid
role	time_stamp	time_stamp	time_stamp	time_stamp	join_key	join_key	join_key	join_key	categorical	categorical	categorical	categorical	categorical	numerical	numerical	numerical	numerical	unused_float	unused_float	unused_float	unused_float	unused_float	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string
unit	time stamp, comparison only	time stamp, comparison only	time stamp, comparison only	time stamp, comparison only
0	2011-05-30 22:00:00	2011-06-12	2011-06-07	2011-06-06 22:00:00	29825	43659	279	5	8	0	279	5	5	20565.6206	1971.5149	616.0984	23153.2339	5	985	985	16281	nan	SO43659	PO522145787	10-4020-000676	105041Vi84182	NULL	79B65321-39CA-4115-9CBA-8FE0903E...
1	2011-05-30 22:00:00	2011-06-12	2011-06-07	2011-06-06 22:00:00	29672	43660	279	5	8	0	279	5	5	1294.2529	124.2483	38.8276	1457.3288	5	921	921	5618	nan	SO43660	PO18850127500	10-4020-000117	115213Vi29411	NULL	738DC42D-D03B-48A1-9822-F95A67EA...
2	2011-05-30 22:00:00	2011-06-12	2011-06-07	2011-06-06 22:00:00	29734	43661	282	6	8	0	282	6	5	32726.4786	3153.7696	985.553	36865.8012	5	517	517	1346	4	SO43661	PO18473189620	10-4020-000442	85274Vi6854	NULL	D91B9131-18A4-4A11-BC3A-90B6F53E...
3	2011-05-30 22:00:00	2011-06-12	2011-06-07	2011-06-06 22:00:00	29994	43662	282	6	8	0	282	6	5	28832.5289	2775.1646	867.2389	32474.9324	5	482	482	10456	4	SO43662	PO18444174044	10-4020-000227	125295Vi53935	NULL	4A1ECFC0-CC3A-4740-B028-1C50BB48...
4	2011-05-30 22:00:00	2011-06-12	2011-06-07	2011-06-06 22:00:00	29565	43663	276	4	8	0	276	4	5	419.4589	40.2681	12.5838	472.3108	5	1073	1073	4322	nan	SO43663	PO18009186470	10-4020-000510	45303Vi22691	NULL	9B1E7A40-6AE0-4AD3-811C-A6495185...
	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
31460	2014-06-29 22:00:00	2014-07-12	2014-07-07	2014-07-06 22:00:00	11981	75119	nan	1	8	1	nan	1	1	42.28	3.3824	1.057	46.7194	5	17649	17649	6761	nan	SO75119	NULL	10-4030-011981	429826Vi35166	NULL	9382F1C9-383A-435F-9449-0EECEA21...
31461	2014-06-29 22:00:00	2014-07-12	2014-07-07	2014-07-06 22:00:00	18749	75120	nan	6	8	1	nan	6	1	84.96	6.7968	2.124	93.8808	5	28374	28374	8925	nan	SO75120	NULL	10-4030-018749	929849Vi46003	NULL	AE6A4FCF-FF73-4CD4-AF2C-5993D00D...
31462	2014-06-29 22:00:00	2014-07-12	2014-07-07	2014-07-06 22:00:00	15251	75121	nan	6	8	1	nan	6	1	74.98	5.9984	1.8745	82.8529	5	26553	26553	14220	nan	SO75121	NULL	10-4030-015251	529864Vi73738	NULL	D7395C0E-00CB-4BFA-A238-0D6A9F49...
31463	2014-06-29 22:00:00	2014-07-12	2014-07-07	2014-07-06 22:00:00	15868	75122	nan	6	8	1	nan	6	1	30.97	2.4776	0.7743	34.2219	5	14616	14616	18719	nan	SO75122	NULL	10-4030-015868	330022Vi97312	NULL	4221035A-4159-492F-AF40-4363A64F...
31464	2014-06-29 22:00:00	2014-07-12	2014-07-07	2014-07-06 22:00:00	18759	75123	nan	6	8	1	nan	6	1	189.97	15.1976	4.7493	209.9169	5	14024	14024	10084	nan	SO75123	NULL	10-4030-018759	230370Vi51970	NULL	D54752FF-2B54-4BE5-95EA-3B72289C...

31465 rows x 28 columns
memory usage: 8.34 MB
name: SalesOrderHeader
type: getml.DataFrame

We must also define customer churn. In this case, we define customer churn as a customer not making another purchase within 180 days of his or her last purchase.

Thus, the churn variable is defined as follows:

0, if another purchase by the same customer has been made within 180 days after OrderDate
1, if no purchase by the same customer has been made within 180 days after OrderDate
NULL, if max(OrderDate) - OrderDate <= 180 days

NULL targets can not be used in our analysis.

In [17]:

  Copied!     
 
sales_order_header_pd = sales_order_header[["OrderDate", "CustomerID", "SalesOrderID"]].to_pandas()

repeat_purchases = sales_order_header_pd.merge(
    sales_order_header_pd[["OrderDate", "CustomerID"]],
    on="CustomerID",
    how="left",
)

repeat_purchases = repeat_purchases[
    repeat_purchases["OrderDate_y"] > repeat_purchases["OrderDate_x"]
]

repeat_purchases = repeat_purchases[
    repeat_purchases["OrderDate_y"] - repeat_purchases["OrderDate_x"] > pd.Timedelta('180 days')
]

repeat_purchases = repeat_purchases.groupby("SalesOrderID", as_index=False).aggregate({"CustomerID": "max"})

repeat_purchase_ids = {sid: True for sid in repeat_purchases["SalesOrderID"]}

cut_off_date = max(sales_order_header_pd["OrderDate"]) - pd.Timedelta('180 days')

churn = np.asarray([
    np.nan if order_date >= cut_off_date else 0 if order_id in repeat_purchase_ids else 1 
    for (order_date, order_id) in zip(sales_order_header_pd["OrderDate"], sales_order_header_pd["SalesOrderID"])
])

sales_order_header["churn"] = churn

sales_order_header = sales_order_header[~sales_order_header.churn.is_nan()].to_df("SalesOrderHeaderRefined")

sales_order_header.set_role("churn", getml.data.roles.target)

sales_order_header
sales_order_header_pd = sales_order_header[["OrderDate", "CustomerID", "SalesOrderID"]].to_pandas() repeat_purchases = sales_order_header_pd.merge( sales_order_header_pd[["OrderDate", "CustomerID"]], on="CustomerID", how="left", ) repeat_purchases = repeat_purchases[ repeat_purchases["OrderDate_y"] > repeat_purchases["OrderDate_x"] ] repeat_purchases = repeat_purchases[ repeat_purchases["OrderDate_y"] - repeat_purchases["OrderDate_x"] > pd.Timedelta('180 days') ] repeat_purchases = repeat_purchases.groupby("SalesOrderID", as_index=False).aggregate({"CustomerID": "max"}) repeat_purchase_ids = {sid: True for sid in repeat_purchases["SalesOrderID"]} cut_off_date = max(sales_order_header_pd["OrderDate"]) - pd.Timedelta('180 days') churn = np.asarray([ np.nan if order_date >= cut_off_date else 0 if order_id in repeat_purchase_ids else 1 for (order_date, order_id) in zip(sales_order_header_pd["OrderDate"], sales_order_header_pd["SalesOrderID"]) ]) sales_order_header["churn"] = churn sales_order_header = sales_order_header[~sales_order_header.churn.is_nan()].to_df("SalesOrderHeaderRefined") sales_order_header.set_role("churn", getml.data.roles.target) sales_order_header

Out[17]:

name	OrderDate	DueDate	ShipDate	ModifiedDate	CustomerID	SalesOrderID	SalesPersonID	TerritoryID	churn	RevisionNumber	OnlineOrderFlag	SalesPersonIDCat	TerritoryIDCat	ShipMethodID	SubTotal	TaxAmt	Freight	TotalDue	Status	BillToAddressID	ShipToAddressID	CreditCardID	CurrencyRateID	SalesOrderNumber	PurchaseOrderNumber	AccountNumber	CreditCardApprovalCode	Comment	rowguid
role	time_stamp	time_stamp	time_stamp	time_stamp	join_key	join_key	join_key	join_key	target	categorical	categorical	categorical	categorical	categorical	numerical	numerical	numerical	numerical	unused_float	unused_float	unused_float	unused_float	unused_float	unused_string	unused_string	unused_string	unused_string	unused_string	unused_string
unit	time stamp, comparison only	time stamp, comparison only	time stamp, comparison only	time stamp, comparison only
0	2011-05-30 22:00:00	2011-06-12	2011-06-07	2011-06-06 22:00:00	29825	43659	279	5	0	8	0	279	5	5	20565.6206	1971.5149	616.0984	23153.2339	5	985	985	16281	nan	SO43659	PO522145787	10-4020-000676	105041Vi84182	NULL	79B65321-39CA-4115-9CBA-8FE0903E...
1	2011-05-30 22:00:00	2011-06-12	2011-06-07	2011-06-06 22:00:00	29672	43660	279	5	0	8	0	279	5	5	1294.2529	124.2483	38.8276	1457.3288	5	921	921	5618	nan	SO43660	PO18850127500	10-4020-000117	115213Vi29411	NULL	738DC42D-D03B-48A1-9822-F95A67EA...
2	2011-05-30 22:00:00	2011-06-12	2011-06-07	2011-06-06 22:00:00	29734	43661	282	6	0	8	0	282	6	5	32726.4786	3153.7696	985.553	36865.8012	5	517	517	1346	4	SO43661	PO18473189620	10-4020-000442	85274Vi6854	NULL	D91B9131-18A4-4A11-BC3A-90B6F53E...
3	2011-05-30 22:00:00	2011-06-12	2011-06-07	2011-06-06 22:00:00	29994	43662	282	6	0	8	0	282	6	5	28832.5289	2775.1646	867.2389	32474.9324	5	482	482	10456	4	SO43662	PO18444174044	10-4020-000227	125295Vi53935	NULL	4A1ECFC0-CC3A-4740-B028-1C50BB48...
4	2011-05-30 22:00:00	2011-06-12	2011-06-07	2011-06-06 22:00:00	29565	43663	276	4	0	8	0	276	4	5	419.4589	40.2681	12.5838	472.3108	5	1073	1073	4322	nan	SO43663	PO18009186470	10-4020-000510	45303Vi22691	NULL	9B1E7A40-6AE0-4AD3-811C-A6495185...
	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
19699	2013-12-30 23:00:00	2014-01-12	2014-01-07	2014-01-06 23:00:00	20826	63358	nan	7	1	8	1	nan	7	1	1173.96	93.9168	29.349	1297.2258	5	24387	24387	3239	nan	SO63358	NULL	10-4030-020826	1142084Vi17039	NULL	41278FBB-3DD8-488B-AEA5-8BF4A6F1...
19700	2013-12-30 23:00:00	2014-01-12	2014-01-07	2014-01-06 23:00:00	24114	63359	nan	10	1	8	1	nan	10	1	1179.47	94.3576	29.4868	1303.3144	5	29682	29682	nan	10770	SO63359	NULL	10-4030-024114	NULL	NULL	749532D4-BFF7-4FEC-9F77-396C4A96...
19701	2013-12-30 23:00:00	2014-01-12	2014-01-07	2014-01-06 23:00:00	11417	63360	nan	7	1	8	1	nan	7	1	548.98	43.9184	13.7245	606.6229	5	21465	21465	6582	nan	SO63360	NULL	10-4030-011417	242387Vi34223	NULL	F92FA2A3-73E1-4DD0-987C-99D1C87E...
19702	2013-12-30 23:00:00	2014-01-12	2014-01-07	2014-01-06 23:00:00	18125	63361	nan	8	1	8	1	nan	8	1	2384.07	190.7256	59.6018	2634.3974	5	26562	26562	11994	nan	SO63361	NULL	10-4030-018125	1242859Vi61993	NULL	3A3758BF-CDFA-4740-9104-87BD5A08...
19703	2013-12-30 23:00:00	2014-01-12	2014-01-07	2014-01-06 23:00:00	15692	63362	nan	10	1	8	1	nan	10	1	2419.06	193.5248	60.4765	2673.0613	5	27090	27090	16083	10770	SO63362	NULL	10-4030-015692	242864Vi83167	NULL	3151D31F-4020-41B2-87F3-B6732869...

19704 rows x 29 columns
memory usage: 5.39 MB
name: SalesOrderHeaderRefined
type: getml.DataFrame

In [18]:

  Copied!     
 
split = getml.data.split.random(train=0.8, test=0.2)
split
split = getml.data.split.random(train=0.8, test=0.2) split

Out[18]:


0	train
1	train
2	train
3	test
4	train
	...

infinite number of rows
type: StringColumnView

In [19]:

  Copied!     
 
container = getml.data.Container(population=sales_order_header, split=split)

container.add(
    product=product,
    sales_order_detail=sales_order_detail, 
    sales_order_header=sales_order_header, 
    sales_order_reason=sales_order_reason,
    special_offer=special_offer,
    store=store,
)

container
container = getml.data.Container(population=sales_order_header, split=split) container.add( product=product, sales_order_detail=sales_order_detail, sales_order_header=sales_order_header, sales_order_reason=sales_order_reason, special_offer=special_offer, store=store, ) container

Out[19]:

population

	subset	name	rows	type
0	test	SalesOrderHeaderRefined	unknown	View
1	train	SalesOrderHeaderRefined	unknown	View

peripheral

	alias	name	rows	type
0	product	Product	504	DataFrame
1	sales_order_detail	SalesOrderDetail	121317	DataFrame
2	sales_order_header	SalesOrderHeaderRefined	19704	DataFrame
3	sales_order_reason	SalesOrderHeaderSalesReason	27647	DataFrame
4	special_offer	SpecialOffer	16	DataFrame
5	store	Store	701	DataFrame

2. Predictive modeling¶

We loaded the data and defined the roles and units. Next, we create a getML pipeline for relational learning.

2.1 Define relational model¶

In [20]:

  Copied!     
 
dm = getml.data.DataModel(sales_order_header.to_placeholder("population"))

dm.add(getml.data.to_placeholder(
    product=product,
    sales_order_detail=sales_order_detail, 
    sales_order_header=sales_order_header, 
    sales_order_reason=sales_order_reason,
    special_offer=special_offer,
    store=store,
))


dm.population.join(
    dm.sales_order_header,
    on="CustomerID",
    time_stamps="OrderDate",
    lagged_targets=True,
    horizon=getml.data.time.days(1),
)

dm.population.join(
    dm.sales_order_detail,
    on="SalesOrderID",
)

dm.population.join(
    dm.sales_order_reason,
    on="SalesOrderID", 
)

dm.population.join(
    dm.store,
    on="SalesPersonID", 
)

dm.sales_order_detail.join(
    dm.product,
    on="ProductID",
    relationship=getml.data.relationship.many_to_one,
)

dm.sales_order_detail.join(
    dm.special_offer,
    on="SpecialOfferID",
    relationship=getml.data.relationship.many_to_one,
)

dm
dm = getml.data.DataModel(sales_order_header.to_placeholder("population")) dm.add(getml.data.to_placeholder( product=product, sales_order_detail=sales_order_detail, sales_order_header=sales_order_header, sales_order_reason=sales_order_reason, special_offer=special_offer, store=store, )) dm.population.join( dm.sales_order_header, on="CustomerID", time_stamps="OrderDate", lagged_targets=True, horizon=getml.data.time.days(1), ) dm.population.join( dm.sales_order_detail, on="SalesOrderID", ) dm.population.join( dm.sales_order_reason, on="SalesOrderID", ) dm.population.join( dm.store, on="SalesPersonID", ) dm.sales_order_detail.join( dm.product, on="ProductID", relationship=getml.data.relationship.many_to_one, ) dm.sales_order_detail.join( dm.special_offer, on="SpecialOfferID", relationship=getml.data.relationship.many_to_one, ) dm 

Out[20]:

diagram

staging

	data frames	staging table
0	population	POPULATION__STAGING_TABLE_1
1	sales_order_detail, product, special_offer	SALES_ORDER_DETAIL__STAGING_TABLE_2
2	sales_order_header	SALES_ORDER_HEADER__STAGING_TABLE_3
3	sales_order_reason	SALES_ORDER_REASON__STAGING_TABLE_4
4	store	STORE__STAGING_TABLE_5

2.2 getML pipeline¶

Set-up the feature learner & predictor

In [21]:

  Copied!     
 
fast_prop = getml.feature_learning.FastProp(
    loss_function=getml.feature_learning.loss_functions.CrossEntropyLoss,
    num_threads=1,    
    num_features=400,
)

predictor = getml.predictors.XGBoostClassifier(n_jobs=1)
fast_prop = getml.feature_learning.FastProp( loss_function=getml.feature_learning.loss_functions.CrossEntropyLoss, num_threads=1, num_features=400, ) predictor = getml.predictors.XGBoostClassifier(n_jobs=1)

Build the pipeline

In [22]:

  Copied!     
 
pipe1 = getml.Pipeline(
    tags=['fast_prop'],
    data_model=dm,
    feature_learners=[fast_prop],
    predictors=[predictor],
    include_categorical=True,
)

pipe1
pipe1 = getml.Pipeline( tags=['fast_prop'], data_model=dm, feature_learners=[fast_prop], predictors=[predictor], include_categorical=True, ) pipe1

Out[22]:

Pipeline(data_model='population',
         feature_learners=['FastProp'],
         feature_selectors=[],
         include_categorical=True,
         loss_function='CrossEntropyLoss',
         peripheral=['product', 'sales_order_detail', 'sales_order_header',
                     'sales_order_reason', 'special_offer', 'store'],
         predictors=['XGBoostClassifier'],
         preprocessors=[],
         share_selected_features=0.5,
         tags=['fast_prop'])

2.3 Model training¶

In [23]:

  Copied!     
 
pipe1.check(container.train)
pipe1.check(container.train)

Checking data model...

  Staging... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:00
  Checking... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:00

The pipeline check generated 2 issues labeled INFO and 0 issues labeled WARNING.

Out[23]:

	type	label	message
0	INFO	FOREIGN KEYS NOT FOUND	When joining POPULATION__STAGING_TABLE_1 and SALES_ORDER_REASON__STAGING_TABLE_4 over 'SalesOrderID' and 'SalesOrderID', there are no corresponding entries for 33.769352% of entries in 'SalesOrderID' in 'POPULATION__STAGING_TABLE_1'. You might want to double-check your join keys.
1	INFO	FOREIGN KEYS NOT FOUND	When joining POPULATION__STAGING_TABLE_1 and STORE__STAGING_TABLE_5 over 'SalesPersonID' and 'SalesPersonID', there are no corresponding entries for 84.941548% of entries in 'SalesPersonID' in 'POPULATION__STAGING_TABLE_1'. You might want to double-check your join keys.

In [24]:

  Copied!     
 
pipe1.fit(container.train)
pipe1.fit(container.train)

Checking data model...

  Staging... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:00

The pipeline check generated 2 issues labeled INFO and 0 issues labeled WARNING.

To see the issues in full, run .check() on the pipeline.

  Staging... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:00
  FastProp: Trying 563 features... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:01
  FastProp: Building features... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:00
  XGBoost: Training as predictor... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:15

Trained pipeline.

Time taken: 0:00:17.669831.

Out[24]:

Pipeline(data_model='population',
         feature_learners=['FastProp'],
         feature_selectors=[],
         include_categorical=True,
         loss_function='CrossEntropyLoss',
         peripheral=['product', 'sales_order_detail', 'sales_order_header',
                     'sales_order_reason', 'special_offer', 'store'],
         predictors=['XGBoostClassifier'],
         preprocessors=[],
         share_selected_features=0.5,
         tags=['fast_prop', 'container-9Cswke'])

2.4 Model evaluation¶

In [25]:

  Copied!     
 
pipe1.score(container.test)
pipe1.score(container.test)

  Staging... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:00
  Preprocessing... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:00
  FastProp: Building features... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:00

Out[25]:

	date time	set used	target	accuracy	auc	cross entropy
0	2024-09-12 12:25:24	train	churn	0.9152	0.9723	0.2187
1	2024-09-12 12:25:25	test	churn	0.9162	0.9699	0.2265

2.5 Features¶

The most important features look as follows:

In [26]:

  Copied!     
 
pipe1.features.to_sql()[pipe1.features.sort(by="importances")[0].name]
pipe1.features.to_sql()[pipe1.features.sort(by="importances")[0].name]

Out[26]:

DROP TABLE IF EXISTS "FEATURE_1_307";

CREATE TABLE "FEATURE_1_307" AS
SELECT MIN( t1."orderdate" - t2."t4__startdate" ) AS "feature_1_307",
       t1.rowid AS rownum
FROM "POPULATION__STAGING_TABLE_1" t1
INNER JOIN "SALES_ORDER_DETAIL__STAGING_TABLE_2" t2
ON t1."salesorderid" = t2."salesorderid"
GROUP BY t1.rowid;

In [27]:

  Copied!     
 
getml.engine.shutdown()
getml.engine.shutdown()

3. Conclusion¶

In this notebook, we successfully demonstrated the process of predicting customer churn for AdventureWorks, a fictional company that sells bicycles, using the getML library. The key steps we covered include:

Background and Data Preparation:
- Introduced AdventureWorks and the purpose of the analysis.
- Loaded and prepared the data from the AdventureWorks database, including various tables such as Product, SalesOrderDetail, SalesOrderHeader, and more.
Data Visualization and Preparation:
- Defined roles for the dataset columns to prepare them for modeling.
- Ensured the data was in the correct format for getML to process.
Predictive Modeling:
- Created a relational model using getML to capture the relationships between different tables.
- Built a getML pipeline for feature learning and prediction.
Model Training and Evaluation:
- Trained the model on the prepared data.
- Evaluated the model's performance to predict customer churn.

By leveraging getML's capabilities, we efficiently handled the complexities of relational data and built a robust classification model to predict customer churn. This approach can be extended to other customer databases and prediction tasks, providing valuable insights and accurate forecasts.

References¶

Motl, Jan, and Oliver Schulte. "The CTU prague relational learning repository." arXiv preprint arXiv:1511.03086 (2015).