Wine Quality

This example uses the Wine Quality dataset to predict the quality of white wine.

• fixed acidity

• volatile acidity

• citric acid

• residual sugar

• total sulfur dioxide

• free sulfur dioxide

• density

• pH

• sulphates

• alcohol

• quality (score between 0 and 10)

We will follow the data science cycle (Data Exploration - Data Preparation - Data Modeling - Model Evaluation - Model Deployment) to solve this problem.

Initialization

This example uses the following version of vastorbit:

import vastorbit as vo

vo.__version__

Connect to VAST. This example uses an existing connection called VASTDSN. For details on how to create a connection, see the Connection tutorial. You can skip the below cell if you already have an established connection.

vo.connect("VASTDSN")

Let’s create a VastFrame of the dataset.

from vastorbit.datasets import load_winequality

winequality = load_winequality()
winequality.head(5)

	123 fixed_acidity Decimal(6,3) 100%	123 volatile_acidity Decimal(7,4) 100%	123 citric_acid Decimal(6,3) 100%	123 residual_sugar Decimal(7,3) 100%	123 chlorides Double 100%	123 free_sulfur_dioxide Decimal(7,2) 100%	123 total_sulfur_dioxide Decimal(7,2) 100%	123 density Double 100%	123 ph Decimal(6,3) 100%	123 sulphates Decimal(6,3) 100%	123 alcohol Double 100%	123 quality Integer 100%	123 good Integer 100%	Abc color Varchar(20) 100%
1	6.3	0.67	0.48	12.6	0.052	57.0	222.0	0.9979	3.17	0.52	9.3	6	0	white
2	7.4	0.4	0.29	5.4	0.044	31.0	122.0	0.994	3.3	0.5	11.1	8	1	white
3	7.1	0.26	0.31	2.2	0.044	29.0	128.0	0.9937	3.34	0.64	10.9	8	1	white
4	9.0	0.31	0.48	6.6	0.043	11.0	73.0	0.9938	2.9	0.38	11.6	5	0	white
5	6.3	0.39	0.24	6.9	0.069	9.0	117.0	0.9942	3.15	0.35	10.2	4	0	white

Data Exploration and Preparation

Let’s explore the data by displaying descriptive statistics of all the columns.

winequality.describe()

	count	mean	std	min	approx_25%	approx_50%	approx_75%	max
"fixed_acidity"	6497.0	7.215	1.296433757799814	3.8	6.422757	6.968065	7.6656013	15.9
"volatile_acidity"	6497.0	0.3397	0.16463647408467857	0.08	0.22848527	0.29178292	0.40539855	1.58
"citric_acid"	6497.0	0.319	0.14531786489759152	0.0	0.24655963	0.31071258	0.3907722	1.66
"residual_sugar"	6497.0	5.443	4.757803743147417	0.6	1.8231287	3.1262202	8.143649	65.8
"chlorides"	6497.0	0.05603386178236098	0.03503360137245909	0.009	0.03823784860061664	0.04706431908012936	0.0648611000668226	0.611
"free_sulfur_dioxide"	6497.0	30.53	17.7493997720025	1.0	17.034159	28.748253	41.335823	289.0
"total_sulfur_dioxide"	6497.0	115.74	56.52185452263028	6.0	77.33673	118.71745	155.37856	440.0
"density"	6497.0	0.9946966338309986	0.002998673003719182	0.98711	0.9923367672715311	0.9948342040334558	0.9969373620697908	1.03898
"ph"	6497.0	3.219	0.16078720210398798	2.72	3.1084259	3.2065077	3.32104	4.01
"sulphates"	6497.0	0.531	0.1488058736144904	0.22	0.4316651	0.5065013	0.6000579	2.0
"alcohol"	6497.0	10.491800831152828	1.192711748870993	8.0	9.485336995965627	10.309458060256224	11.308362700676847	14.9
"quality"	6497.0	5.818377712790519	0.8732552715311251	3.0	5.0	6.0	6.0	9.0
"good"	6497.0	0.19655225488687086	0.39742140889536703	0.0	0.0	0.0	0.0	1.0

The quality of a wine is based on the equilibrium between certain components:

• For red wines: tannin/smoothness/acidity

• For white wines: smoothness/acidity

Based on this, we don’t have the data to create a good model for red wines (the tannins weren’t extracted). We do, however, have enough data to make a good model for white wines, so let’s filter out red wines from our study.

winequality.filter(winequality["color"] == "white").drop(["good", "color"])

	123 fixed_acidity Decimal(6,3) 100%	123 volatile_acidity Decimal(7,4) 100%	123 citric_acid Decimal(6,3) 100%	123 residual_sugar Decimal(7,3) 100%	123 chlorides Double 100%	123 free_sulfur_dioxide Decimal(7,2) 100%	123 total_sulfur_dioxide Decimal(7,2) 100%	123 density Double 100%	123 ph Decimal(6,3) 100%	123 sulphates Decimal(6,3) 100%	123 alcohol Double 100%	123 quality Integer 100%
1	8.8	0.19	0.3	5.0	0.028	34.0	120.0	0.99242	2.94	0.47	11.2	5
2	7.7	0.46	0.18	3.3	0.054	18.0	143.0	0.99392	3.12	0.51	10.8	6
3	8.8	0.27	0.25	5.0	0.024	52.0	99.0	0.9925	2.87	0.49	11.4	5
4	5.8	0.18	0.28	1.3	0.034	9.0	94.0	0.99092	3.21	0.52	11.2	6
5	5.8	0.15	0.32	1.2	0.037	14.0	119.0	0.99137	3.19	0.5	10.2	6
6	5.6	0.32	0.33	7.4	0.037	25.0	95.0	0.99268	3.25	0.49	11.1	6
7	6.1	0.33	0.32	7.8	0.052	52.0	183.0	0.99657	3.39	0.65	9.5	5
8	7.1	0.32	0.3	9.9	0.041	63.0	192.0	0.99642	3.12	0.49	10.2	6
9	6.2	0.23	0.35	0.7	0.051	24.0	111.0	0.9916	3.37	0.43	11.0	3
10	8.9	0.3	0.35	4.6	0.032	32.0	148.0	0.99458	3.15	0.45	11.5	7
11	6.0	0.14	0.17	5.6	0.036	37.0	127.0	0.99373	3.05	0.57	9.8	6
12	6.8	0.24	0.29	9.5	0.042	56.0	157.0	0.99586	3.11	0.51	10.1	6
13	6.7	0.21	0.48	14.8	0.05	31.0	195.0	0.99942	2.95	0.75	8.8	6
14	8.9	0.3	0.35	4.6	0.032	32.0	148.0	0.99458	3.15	0.45	11.5	7
15	6.1	0.3	0.3	2.1	0.031	50.0	163.0	0.9895	3.39	0.43	12.7	7
16	7.2	0.37	0.4	11.6	0.032	34.0	214.0	0.9963	3.1	0.51	9.8	6
17	6.7	0.64	0.3	1.2	0.03	18.0	76.0	0.9892	3.16	0.6	12.9	4
18	7.2	0.37	0.4	11.6	0.032	34.0	214.0	0.9963	3.1	0.51	9.8	6
19	6.1	0.3	0.3	2.1	0.031	50.0	163.0	0.9895	3.39	0.43	12.7	7
20	7.6	0.28	0.49	20.15	0.06	30.0	145.0	1.00196	3.01	0.44	8.5	5

Let’s draw the correlation matrix of the dataset.

winequality.corr(method = "spearman")

We can see a strong correlation between the density and the alcohol degree (the alcohol degree describes the density of pure ethanol in the wine).

We can drop the density column since it doesn’t influence the quality of the white wine (instead, its presence will just bias the data).

winequality.drop(["density"])

	123 fixed_acidity Decimal(6,3) 100%	123 volatile_acidity Decimal(7,4) 100%	123 citric_acid Decimal(6,3) 100%	123 residual_sugar Decimal(7,3) 100%	123 chlorides Double 100%	123 free_sulfur_dioxide Decimal(7,2) 100%	123 total_sulfur_dioxide Decimal(7,2) 100%	123 ph Decimal(6,3) 100%	123 sulphates Decimal(6,3) 100%	123 alcohol Double 100%	123 quality Integer 100%
1	6.3	0.67	0.48	12.6	0.052	57.0	222.0	3.17	0.52	9.3	6
2	7.4	0.4	0.29	5.4	0.044	31.0	122.0	3.3	0.5	11.1	8
3	7.1	0.26	0.31	2.2	0.044	29.0	128.0	3.34	0.64	10.9	8
4	9.0	0.31	0.48	6.6	0.043	11.0	73.0	2.9	0.38	11.6	5
5	6.3	0.39	0.24	6.9	0.069	9.0	117.0	3.15	0.35	10.2	4
6	8.2	0.22	0.36	6.8	0.034	12.0	90.0	3.01	0.38	10.5	8
7	7.1	0.19	0.28	3.6	0.033	16.0	78.0	2.91	0.78	11.4	6
8	7.3	0.25	0.36	13.1	0.05	35.0	200.0	3.04	0.46	8.9	7
9	7.9	0.2	0.34	1.2	0.04	29.0	118.0	3.14	0.41	10.6	6
10	7.1	0.26	0.32	5.9	0.037	39.0	97.0	3.31	0.4	11.6	6
11	7.0	0.2	0.34	5.7	0.035	32.0	83.0	3.19	0.46	11.5	6
12	6.9	0.3	0.33	4.1	0.035	26.0	155.0	3.25	0.79	12.3	8
13	8.1	0.29	0.49	7.1	0.042	22.0	124.0	3.14	0.41	10.8	6
14	5.8	0.17	0.3	1.4	0.037	55.0	130.0	3.29	0.38	11.3	6
15	5.9	0.415	0.02	0.8	0.038	22.0	63.0	3.36	0.36	9.3	5
16	6.6	0.23	0.26	1.3	0.045	16.0	128.0	3.36	0.6	10.0	6
17	8.6	0.55	0.35	15.55	0.057	35.5	366.5	3.04	0.63	11.0	3
18	6.9	0.35	0.74	1.0	0.044	18.0	132.0	3.13	0.55	10.2	5
19	7.6	0.14	0.74	1.6	0.04	27.0	103.0	3.07	0.4	10.8	7
20	9.2	0.28	0.49	11.8	0.042	29.0	137.0	3.1	0.34	10.1	4

We’re working with the scores given by wine tasters, so it’s likely that two closely competing wines will have a similar score. Knowing this, a k-nearest neighbors (KNN) model would be best.

KNN is sensitive to unnormalized data so we’ll have to normalize our data.

winequality.normalize(
    [
        "free_sulfur_dioxide",
        "residual_sugar",
        "pH",
        "sulphates",
        "volatile_acidity",
        "fixed_acidity",
        "citric_acid",
        "chlorides",
        "total_sulfur_dioxide",
        "alcohol"
    ],
    method = "robust_zscore",
)

	123 fixed_acidity Real 100%	123 volatile_acidity Real 100%	123 citric_acid Real 100%	123 residual_sugar Real 100%	123 chlorides Real 100%	123 free_sulfur_dioxide Real 100%	123 total_sulfur_dioxide Real 100%	123 ph Real 100%	123 sulphates Real 100%	123 alcohol Real 100%	123 quality Integer 100%
1	-0.6263393225607145	4.911155783031988	1.9605633466538805	1.336724760954758	0.8425373442974716	1.4456305327191328	2.016276711787808	-0.046518225450507766	0.4414422660014937	-0.7560787364264695	6
2	0.7791104333461998	1.6587446267190056	-0.2999106476710552	0.021945258586426153	0.10323904098655225	-0.15602509850974183	-0.2838021278673736	0.8603785845680894	0.2492828864934583	0.5384354479461952	8
3	0.3958059544624959	-0.02769078766550382	-0.06196601668948299	-0.5624011869106101	0.10323904098655225	-0.27922937783503987	-0.1457973974880627	1.1394237568815038	1.594398543049706	0.3946005385714551	8
4	2.8234009873926205	0.574607574614678	1.9605633466538805	0.24107517564781478	0.01082675307268724	-1.3880678917627223	-1.4108407592984127	-1.9300731385660557	-0.9036733905547539	0.8980227213830467	5
5	-0.6263393225607145	1.538284954262969	-0.8947722251249857	0.2958576549131619	2.413546238833176	-1.5112721710880204	-0.3988060698501327	-0.186040811607215	-1.1919124598168072	-0.10882164424013777	4
6	1.8012557103694102	-0.5095294774896494	0.5328955607644474	0.27759682849137957	-0.8208838381520965	-1.3264657521000733	-1.0198273565570317	-1.1626989147041658	-0.9036733905547539	0.10693071982197365	8
7	0.3958059544624959	-0.8709084948577585	-0.41888296316184126	-0.3067496170056568	-0.9132961260659616	-1.0800571934494771	-1.2958368173156536	-1.860311845487702	2.9395141996059535	0.7541878120083066	6
8	0.6513422737182984	-0.1481504601215402	0.5328955607644474	1.4280288930636698	0.6577127684697422	0.09038346014085426	1.5102593670636681	-0.9534150354691049	-0.13503587252261245	-1.0437485551759509	7
9	1.4179512314857063	-0.7504488224017221	0.2949509297828753	-0.7450094511284341	-0.26641011066890713	-0.27922937783503987	-0.3758052814535808	-0.25580210468556863	-0.6154343212927009	0.1788481745093437	6
10	0.3958059544624959	-0.02769078766550382	0.0570062988013031	0.11324939069533808	-0.5436469744105021	0.3367920187914504	-0.858821837781169	0.930139877646443	-0.7115140110467186	0.8980227213830467	6
11	0.2680377948345946	-0.7504488224017221	0.2949509297828753	0.07672773785177331	-0.7284715502382315	-0.09442295884709281	-1.1808328753328945	0.09300436070619948	-0.13503587252261245	0.8261052666956766	6
12	0.1402696352066933	0.4541479021586417	0.1759786142920892	-0.21544548489674487	-0.7284715502382315	-0.46403579682298696	0.4752238892188364	0.5115721191763213	3.0355938893599714	1.4014449041946395	8
13	1.6734875507415088	0.3336882297026053	2.0795356621446666	0.3323793077567267	-0.08158553484117712	-0.7104443554735831	-0.2378005510742699	-0.25580210468556863	-0.6154343212927009	0.32268308388408506	6
14	-1.265180120700221	-1.1118278397698313	-0.18093833218026908	-0.7084877982848693	-0.5436469744105021	1.3224262533938347	-0.099795820694959	0.7906172914897357	-0.9036733905547539	0.6822703573209365	6
15	-1.1374119610723197	1.83943413540306	-3.5121631659222796	-0.8180527568155636	-0.4512346864966371	-0.7104443554735831	-1.6408486432639309	1.278946343038211	-1.0958327700627895	-0.7560787364264695	5
16	-0.2430348436770106	-0.38906980503361294	-0.6568275941434134	-0.7267486247066517	0.19565132890041725	-1.0800571934494771	-0.1457973974880627	1.278946343038211	1.2100797840336353	-0.2526565536148779	6
17	2.3123283488810156	3.4656397135595514	0.41392324527366137	1.8754191403973384	1.3045987838667967	0.12118452997217878	5.339890635089546	-0.9534150354691049	1.4983188532956884	0.46651799325882515	3
18	0.1402696352066933	1.0564462644388235	5.053843549414319	-0.7815311039719988	0.10323904098655225	-0.9568529141241792	-0.0537942439018554	-0.32556339776392224	0.7296813352635468	-0.10882164424013777	5
19	1.0346467526020025	-1.4732068571379404	5.053843549414319	-0.6719661454413045	-0.26641011066890713	-0.4024336571603379	-0.7208171074018581	-0.744131156234044	-0.7115140110467186	0.32268308388408506	7
20	3.0789373066484234	0.21322855724656894	2.0795356621446666	1.1906381495804987	-0.08158553484117712	-0.27922937783503987	0.0612096980809037	-0.5348472769989832	-1.2879921495708246	-0.1807390989275078	4

Machine Learning

Let’s create our KNN model.

from vastorbit.machine_learning.vast import KNeighborsRegressor
from vastorbit.machine_learning.model_selection import cross_validate

predictors = winequality.get_columns(exclude_columns = ["quality"])
model = KNeighborsRegressor(name = "winequality_KNN", n_neighbors = 50)
cross_validate(model, winequality, predictors, "quality")

	explained_variance	max_error	median_absolute_error	mean_absolute_error	mean_squared_error	root_mean_squared_error	r2	r2_adj	aic	bic	time
1-fold	7.771561172376096e-16	2.62	0.6200000000000001	0.7688944099378929	0.8388720496894408	0.9339703534470446	-0.08851703309372083	-0.09532451922939145	-260.5556650591127	-201.6484268286082	0.0002968311309814453
2-fold	-2.220446049250313e-16	3.34	0.6572532188841202	0.8502504581551539	0.8313092241905929	1.0303272991195274	-0.2889521080685411	-0.29687924280450995	-280.1300005990673	-221.0345588633694	0.0028929710388183594
3-fold	1.1102230246251565e-15	3.0	0.7762824850261885	0.6267308850090307	0.791559060806743	0.8951673046482754	-0.015724310680232723	-0.021880215593446106	-365.9531989405429	-306.693125372759	0.00356292724609375
avg	5.551115123125783e-16	2.986666666666667	0.6845119013034363	0.7486252510340258	0.820580111562259	0.9531549857382825	-0.13106448394749823	-0.13802799254244916	-302.212954866241	-243.1253703549122	0.0022509098052978516
std	5.661048867003676e-16	0.29408993333483696	0.06664998428065853	0.0923701935938895	0.020751949669001866	0.056821906652743934	0.1155308659970214	0.11625777742157621	45.77410249581722	45.64062618124369	0.001408551958331754

Our model is good enough for the use case. Our predicted scores have a median absolute error of less than 1. If we want to improve this model, we’ll probably need more relevant features.

Conclusion

We’ve solved our problem in a pandas-like way, all without ever loading data into memory!