Distributed Logistic Model Trees, Stratio Intelligence Mateo Álvarez and Antonio

Transcription

1 Distributed Logistic Model Trees, Stratio Intelligence Mateo Álvarez and Antonio

2 Aerospace Engineer, MSc in Propulsion Systems (UPM), Master in Data Science (URJC). Working as data scientist and Big Data developer at Stratio Big Data in the data science department mateo-alvarez

3 Ph.D. in Telecommunications, MSc in Electronic Systems Engineering and Telecommunication Technologies, Systems and Networks (UPV), and MSc Big Data Expert (UTAD). Working as data scientist and Big Data developer at at Stratio Big Data in the data science

4 Why using interpretable algorithms instead of black boxes Logistic Regression Metrics Demo Decision Trees Variance-Bias tradeoff Logistic Model Trees Distributed implementation Cost function & configuration params Demo

5 Why use interpretable algorithms instead of black boxes Logistic Regression Decision Trees Variance-Bias

6 Accuracy VS Explainability Medical Studies Power management Financial environment Criminal activity

7 Probability Threshold Feature

8 Probability Local bad adjust Threshold Feature

9 Probability Threshold Feature

10 Root node Feature 1 Leaf node Leaf node

11 Root node Feature 1 Feature 2 Feature 2 Leaf node Leaf node Leaf node Leaf node

12 Root node Feature 1 Local bad adjust Feature 2 Feature 2 Leaf node Leaf node Leaf node Leaf node

13 Mean Error Error Total error Test error Underfitting Overfitting Training error Variance Bias 2 Model complexity Model complexity

14 Variance Bias

15 Missing important variables for the problem to make the predictions Bias Variance

16 Overfitting to the sample/training data Bias Variance

17 Irreducible error on prediction Bias Variance

18 Logistic Model Trees Distributed implementation Cost function & configuration parametres

19 Performance metrics Root node

20 Root node Feature 1 Performance metrics Performance metrics Performance metrics

21 Root node Feature 1 Performance metrics Performance metrics Performance metrics Performance metrics Performance metrics Performance metrics

22 Root node Feature 1 Feature 2 Feature 2

23 Root node Feature 1 Feature 2

24 Root node Feature 1 Feature 2

25 DISTRIBUTED IMPLEMENTATION Spark s Decision Tree (distributed implementation of random forests) Spark s Logistic Regression / weka s Logistic Regression on the nodes

26 LMT Cost function to fix the logistic regression threshold AccuracyCostFunction ConfusionMatrix PrecisionCostFunction PrecisionRecallCostFunction RocCostFunction The same cost function for pruning criteria Performance metrics Performance metrics Performance metrics

27 ADVANTAGES OF THIS IMPLEMENTATION Big datasets Power of spark to distribute building the tree and logistic regressions Medium datasets Distributed tree growth and weka s logistic regression Small datasets Although it can be slow to distribute the data for the decision tree, cost functions can be still used and specific optimization for particular cases

28 Example of DLMT algorithm in a synthetic dataset

29 Metrics

30 PREDICTION Positive Negative TRUE CONDITION Positive True Positives False Negatives Negative False Positives True Negatives True Positive Rate (Recall) False Positive Rate Precission TPR = TP/(TP+FN) Insensitive to unbalance FPR = FP/(FP+TN) Insensitive to unbalance Precision = TP/(TP+FP) Sensitive to unbalance Accuracy = (TP+TN)/(TP+TN+FP+FN) Sensitive to unbalance

31 PREDICTION Positive Negative TRUE CONDITION Positive True Positives False Negatives Negative False Positives True Negatives True Positive Rate (Recall) False Positive Rate Best performance AUROC (AUC): TPR/FPR -> Insensitive to unbalance! TPR FPR

32 PREDICTION Positive Negative TRUE CONDITION Positive True Positives False Negatives Negative False Positives True Negatives True Positive Rate (Recall) Precission Best performance AUPRC: Precision/TPR -> Sensitive to unbalance! Precision Recall

33 Data Algorithms f f 1 n ABF Benchmark

34 @StratioBD

35 1 Accuracy VS Explainability 2 3 Performance Metrics: 4 AUROC, AUPRC, ACCURACY Automatic Benchmarking Framework f f 1 n ABF Benchmark

36 THANK YOU UNITED STATES Tel: (+1) EUROPE Tel: (+34)

37 WE ARE