Machine Learning using H2O.ai

Import Library

H2O the leading open source machine learning and artificial intelligence platform. This leading platform undoubtedly makes scripting of algorithms seemingly easy. It typically uses driverless technology that promptly removes the specific need to invariably do extensive and costly feature engineering upfront, in addition to automating model validation and tuning. You do not need extensive knowledge of programming to properly understand the platform. It is quite simple and straightforward. It graciously allows organisations to unanimously adopt AI. 

Underneath, the H2O library was imported. This library gallantly helped us create arrays and profile tables, compute graphs and make accurate predictions.

import h2o
h2o.init()

Checking whether there is an H2O instance running at http://localhost:54321 . connected.

H2O_cluster_uptime:	3 hours 26 mins
H2O_cluster_timezone:	Africa/Harare
H2O_data_parsing_timezone:	UTC
H2O_cluster_version:	3.30.0.4
H2O_cluster_version_age:	17 days
H2O_cluster_name:	H2O_from_python_Tshepo_j6l31n
H2O_cluster_total_nodes:	1
H2O_cluster_free_memory:	1.972 Gb
H2O_cluster_total_cores:	8
H2O_cluster_allowed_cores:	8
H2O_cluster_status:	locked, healthy
H2O_connection_url:	http://localhost:54321
H2O_connection_proxy:	{“http”: null, “https”: null}
H2O_internal_security:	False
H2O_API_Extensions:	Amazon S3, Algos, AutoML, Core V3, TargetEncoder, Core V4
Python_version:	3.7.4 final

Load data

df = h2o.import_file(r"C:\Users\Tshepo\Desktop\MLAgortihms\Datasets\diabetes.csv")

The h2o.import_file() function was called to load obtained csv file into a pandas data frame. See underneath table to examine the data structure. Bear in mind that time was parsed by calling the parse_dates function and then turned into an index

Parse progress: |█████████████████████████████████████████████████████████| 100%

df.col_names

Assigning x and y

Below are the features that were used as independent variables.

[‘Pregnancies’, ‘Glucose’, ‘BloodPressure’, ‘SkinThickness’, ‘Insulin’, ‘BMI’, ‘DiabetesPedigreeFunction’, ‘Age’]

Outcome was used as a dependent variable.

y = "Outcome"
x = df.col_names
x.remove("Outcome")

df["Outcome"] = df["Outcome"].asfactor()

df["Outcome"].levels()

[['0', '1']]

Split data into training, validation and test data

The data was split into training data and test data using the code underneath. The training data was used to fit the data. Meanwhile, the test data was conveniently used to validate our classification model.

train, valid, test = df.split_frame(ratios=[.8, .1], seed=1234)
print("Train shape: ", train.shape)
print("Valid shape: ", valid.shape)
print("Test shape: ", test.shape)

The shape of the training, validation and test data is as follows.

Train shape:  (613, 9)
Valid shape:  (69, 9)
Test shape:  (86, 9)

Configuring logistic regression model

The code underneath finalizes logistic regression model.

from h2o.estimators.glm import H2OGeneralizedLinearEstimator
glm = H2OGeneralizedLinearEstimator(family="binomial")
glm.train(x=x,y=y,training_frame=train,validation_frame=valid)

glm Model Build progress: |███████████████████████████████████████████████| 100%

Predictions

The predict() function was called to predict salaries. Thereafter, a data frame consisting of predicted values was created.

y_predglm = glm.predict(test)
y_predglm

glm prediction progress: |████████████████████████████████████████████████| 100%

predict	p0	p1
0	0.956033	0.0439669
0	0.977438	0.0225621
1	0.58329	0.41671
0	0.753834	0.246166
1	0.269132	0.730868
0	0.952432	0.0475679
0	0.811075	0.188925
0	0.917553	0.0824472
0	0.665598	0.334402
0	0.813001	0.186999

Actual Values

test

We compute actual values to compare them with predicted values.

Pregnancies	Glucose	BloodPressure	SkinThickness	Insulin	BMI	DiabetesPedigreeFunction	Age	Outcome
1	89	66	23	94	28.1	0.167	21	0
8	125	96	0	0	0	0.232	54	1
1	103	30	38	83	43.3	0.183	33	0
1	115	70	30	96	34.6	0.529	32	1
7	147	76	0	0	39.4	0.257	43	1
3	88	58	11	54	24.8	0.267	22	0
2	90	68	42	0	38.2	0.503	27	1
1	101	50	15	36	24.2	0.526	26	0
7	114	66	0	0	32.8	0.258	42	1
0	109	88	30	0	32.5	0.855	38	1

Model evaluation

The section underneath outlines the area under the curve for the training, validation and test data.

train_modelevaluation = glm.model_performance(test_data=train)
valid_modelevaluation = glm.model_performance(test_data=valid)
test_modelevaluation = glm.model_performance(test_data=test)
print("Train AUC: ", train_modelevaluation.auc())
print("Valid AUC:", valid_modelevaluation.auc())
print("Test AUC:", test_modelevaluation.auc())

Train AUC: 0.8414036680165713

Valid AUC: 0.774074074074074

Test AUC: 0.855

The section underneath outlines the gini coefficient for the training, validation and test data.

print("Train Gini: ", train_modelevaluation.gini())
print("Valid Gini:", valid_modelevaluation.gini())
print("Test Gini:", test_modelevaluation.gini())

Train Gini: 0.6828073360331426

Valid Gini: 0.548148148148148 Test Gini: 0.71

Test Gini: 0.71

print(train_modelevaluation)

Model Evaluation

In the next section we evaluated the performance of the logistic regression model.

ModelMetricsBinomialGLM: glm
** Reported on test data. **

MSE: 0.15226071742967404
RMSE: 0.3902059935850217
LogLoss: 0.47072042475895653
Null degrees of freedom: 612
Residual degrees of freedom: 604
Null deviance: 796.7599047076832
Residual deviance: 577.1032407544807
AIC: 595.1032407544807
AUC: 0.8414036680165713
AUCPR: 0.7356845220146738
Gini: 0.6828073360331426

Confusion Matrix (Act/Pred) for max f1 @ threshold = 0.3356601829569534:

0	1	Error	Rate
0	0	306.0	90.0	0.2273	(90.0/396.0)
1	1	49.0	168.0	0.2258	(49.0/217.0)
2	Total	355.0	258.0	0.2268	(139.0/613.0)

Maximum Metrics: Maximum metrics at their respective thresholds

	metric	threshold	value	idx
0	max f1	0.335660	0.707368	195.0
1	max f2	0.140083	0.805215	306.0
2	max f0point5	0.533112	0.715922	127.0
3	max accuracy	0.533112	0.786297	127.0
4	max precision	0.993830	1.000000	0.0
5	max recall	0.014357	1.000000	393.0
6	max specificity	0.993830	1.000000	0.0
7	max absolute_mcc	0.335660	0.529758	195.0
8	max min_per_class_accuracy	0.335660	0.772727	195.0
9	max mean_per_class_accuracy	0.335660	0.773460	195.0
10	max tns	0.993830	396.000000	0.0
11	max fns	0.993830	216.000000	0.0
12	max fps	0.002006	396.000000	399.0
13	max tps	0.014357	217.000000	393.0
14	max tnr	0.993830	1.000000	0.0
15	max fnr	0.993830	0.995392	0.0
16	max fpr	0.002006	1.000000	399.0
17	max tpr	0.014357	1.000000	393.0

Gains/Lift Table: Avg response rate: 35.40 %, avg score: 35.40 %

		group	cumulative_data_fraction	lower_threshold	lift	cumulative_lift	response_rate	score	cumulative_response_rate	cumulative_score	capture_rate	cumulative_capture_rate	gain	cumulative_gain
0		1	0.011419	0.959349	2.421330	2.421330	0.857143	0.969492	0.857143	0.969492	0.027650	0.027650	142.132982	142.132982
1		2	0.021207	0.928114	1.883257	2.172988	0.666667	0.943412	0.769231	0.957455	0.018433	0.046083	88.325653	117.298830
2		3	0.030995	0.911829	2.354071	2.230172	0.833333	0.919732	0.789474	0.945542	0.023041	0.069124	135.407066	123.017220
3		4	0.040783	0.896808	2.824885	2.372903	1.000000	0.904451	0.840000	0.935680	0.027650	0.096774	182.488479	137.290323
4		5	0.050571	0.875446	2.354071	2.369258	0.833333	0.882663	0.838710	0.925419	0.023041	0.119816	135.407066	136.925821
5		6	0.101142	0.795272	2.551509	2.460384	0.903226	0.836540	0.870968	0.880979	0.129032	0.248848	155.150884	146.038353
6		7	0.150082	0.720283	2.259908	2.395011	0.800000	0.751225	0.847826	0.838668	0.110599	0.359447	125.990783	139.501102
7		8	0.200653	0.643244	1.913632	2.273688	0.677419	0.686057	0.804878	0.800205	0.096774	0.456221	91.363163	127.368776
8		9	0.300163	0.466522	1.528216	2.026548	0.540984	0.563476	0.717391	0.721724	0.152074	0.608295	52.821636	102.654779
9		10	0.399674	0.355216	1.204049	1.821762	0.426230	0.404886	0.644898	0.642838	0.119816	0.728111	20.404926	82.176244
10		11	0.500816	0.277274	0.911253	1.637881	0.322581	0.314918	0.579805	0.576613	0.092166	0.820276	-8.874684	63.788108
11		12	0.600326	0.210118	0.787263	1.496882	0.278689	0.241369	0.529891	0.521043	0.078341	0.898618	-21.273703	49.688189
12		13	0.699837	0.146909	0.602025	1.369641	0.213115	0.178318	0.484848	0.472310	0.059908	0.958525	-39.797537	36.964111
13		14	0.799347	0.099871	0.185238	1.222195	0.065574	0.120213	0.432653	0.428478	0.018433	0.976959	-81.476165	22.219505
14		15	0.898858	0.059240	0.092619	1.097142	0.032787	0.081178	0.388385	0.390029	0.009217	0.986175	-90.738083	9.714219
15		16	1.000000	0.001990	0.136688	1.000000	0.048387	0.033778	0.353997	0.353997	0.013825	1.000000	-86.331203	0.000000