ML Demo Bay

Machine Learning

Three tiny models on a real data excerpt.

This page keeps machine learning small on purpose: the UCI Wine Quality data, a cached training artifact, and instant predictions without runtime training. The buttons replay saved checkpoints; the browser is not fitting models.

dataset.csv

Dataset

UCI Wine Quality

Holdout test split

Full UCI red wine dataset with 1599 rows and 11 chemical features. Visible examples come from the holdout test split; for classification, quality >= 7 is good.

Samples: 24
Variant: red
License: CC BY 4.0
Split: Holdout test split
Training: 1119
Validation: 240
Test: 240

Sample

Actual quality: 5 / 10
Actual class: standard

Feature values

fixed acidity 7.3
volatile acidity 1.07
citric acid 0.09
residual sugar 1.7
chlorides 0.178
free sulfur dioxide 10
total sulfur dioxide 89
density 0.9962
pH 3.3
sulphates 0.57
alcohol 9

Source and attribution

Cortez, Cerdeira, Almeida, Matos, and Reis, 2009.

UCI Machine Learning Repository winequality-red.csv

Regression

Linear Regression

Estimates wine quality as a number from 0 to 10.

Target: quality

Prediction

4.24 / 10 Absolute error: 0.76

Cached training replay

Final model: cached ridge weights

Checkpoint: 5
Selected features: 11
Progress: 100%

Saved metrics No browser training Final model: Standardized on train split 11

fixed acidity +0.08
volatile acidity -0.21
citric acid -0.06
residual sugar +0.03
chlorides -0.11
free sulfur dioxide +0.04
total sulfur dioxide -0.1
density -0.08
pH -0.06
sulphates +0.18
alcohol +0.26

Metrics

Test Accuracy 85% Test F1 0.42 Test Recall 34% Test Precision 54% Balanced Acc 64% Accuracy +/-1 90% Test MAE 0.51

Feature weights

fixed acidity +0.08
volatile acidity -0.21
citric acid -0.06
residual sugar +0.03
chlorides -0.11
free sulfur dioxide +0.04
total sulfur dioxide -0.1
density -0.08
pH -0.06
sulphates +0.18
alcohol +0.26

Classification

Logistic Regression

Estimates whether a sample belongs to the good class.

Target: quality >= 7

Prediction

0% standard Probability good: 0.003

Cached training replay

Cached epoch 900 checkpoint

Checkpoint: 900
Validation F1: 0.538
Progress: 100%

Saved metrics No browser training Final model: Validation accuracy 0.8

fixed acidity +0.5
volatile acidity -0.43
citric acid +0.02
residual sugar +0.37
chlorides -0.39
free sulfur dioxide +0.07
total sulfur dioxide -0.47
density -0.45
pH -0.007
sulphates +0.71
alcohol +1.04

Metrics

Test Accuracy 85% Test F1 0.41 Test Recall 32% Test Precision 57% Balanced Acc 64%

Feature weights

fixed acidity +0.5
volatile acidity -0.43
citric acid +0.02
residual sugar +0.37
chlorides -0.39
free sulfur dioxide +0.07
total sulfur dioxide -0.47
density -0.45
pH -0.007
sulphates +0.71
alcohol +1.04

Confusion Matrix

FN: 26
FP: 9
TN: 193
TP: 12

Classification

Decision Tree

Shows a small, readable rule path.

Target: quality >= 7

Prediction

5 / 10 -> standard Rule path

Cached training replay

Final model: cached CART rules

Checkpoint: 9
Validation F1: 0.583
Progress: 100%

Saved metrics No browser training Final model: Balanced accuracy 0.775

alcohol > 10.45
sulphates > 0.635
alcohol > 11.15
alcohol > 11.65

FN: 12
FP: 18
TN: 189
TP: 21

Metrics

Test Accuracy 86% Test F1 0.51 Test Recall 47% Test Precision 56% Balanced Acc 70%

Tree rules

alcohol > 10.45
sulphates > 0.635
alcohol > 11.15
alcohol > 11.65

Rule path

alcohol <= 10.45
fixed_acidity <= 11.45
sulphates <= 0.595
citric_acid > 0.025

Confusion Matrix

FN: 20
FP: 14
TN: 188
TP: 18