Kalman Filters and Adaptive Windows for Learning
                 in Data Streams

                             Albert Bif...
Outline

1   Introduction

2   The Kalman Filter and the CUSUM Test

3   The ADWIN Algorithm

4   General Framework

5   K...
Introduction


Introduction


    Data Streams
            Sequence potentially infinite
            High amount of data:
 ...
Introduction


Introduction




Problem
Given an input sequence x1 , x2 , . . . , xt , . . . we want to output at instant ...
Introduction


Time Change Detectors and Predictors: A General
Framework


                                               ...
Introduction


Time Change Detectors and Predictors: A General
Framework


                                               ...
Introduction


Time Change Detectors and Predictors: A General
Framework


                                               ...
Introduction


Introduction



                                                 Our particular proposal: K-ADWIN
  Our gen...
Introduction


Data Mining Algorithms with Concept Drift

 No Concept Drift                                      Concept d...
Introduction


Data Mining Algorithms with Concept Drift

 No Concept Drift                                      Concept D...
The Kalman Filter and the CUSUM Test


The Kalman Filter
    Optimal recursive algorithm
    Minimum mean-square error est...
The Kalman Filter and the CUSUM Test


The Kalman Filter



The difference equation of our discrete-time controlled proces...
The Kalman Filter and the CUSUM Test


The Kalman Filter


The difference equation of our discrete-time controlled process...
The Kalman Filter and the CUSUM Test


The CUSUM Test



    The cumulative sum (CUSUM algorithm),is a change detection
  ...
The ADWIN Algorithm


Algorithm ADWIN [BG06]

Example
W = 101010110111111



ADWIN: A DAPTIVE W INDOWING A LGORITHM
1 Init...
The ADWIN Algorithm


Algorithm ADWIN [BG06]
Example
W = 101010110111111
W0 = 1 W1 = 01010110111111



ADWIN: A DAPTIVE W ...
The ADWIN Algorithm


Algorithm ADWIN [BG06]
Example
W = 101010110111111
W0 = 10 W1 = 1010110111111



ADWIN: A DAPTIVE W ...
The ADWIN Algorithm


Algorithm ADWIN [BG06]
Example
W = 101010110111111
W0 = 101 W1 = 010110111111



ADWIN: A DAPTIVE W ...
The ADWIN Algorithm


Algorithm ADWIN [BG06]
Example
W = 101010110111111
W0 = 1010 W1 = 10110111111



ADWIN: A DAPTIVE W ...
The ADWIN Algorithm


Algorithm ADWIN [BG06]
Example
W = 101010110111111
W0 = 10101 W1 = 0110111111



ADWIN: A DAPTIVE W ...
The ADWIN Algorithm


Algorithm ADWIN [BG06]
Example
W = 101010110111111
W0 = 101010 W1 = 110111111



ADWIN: A DAPTIVE W ...
The ADWIN Algorithm


Algorithm ADWIN [BG06]
Example
W = 101010110111111
W0 = 1010101 W1 = 10111111



ADWIN: A DAPTIVE W ...
The ADWIN Algorithm


Algorithm ADWIN [BG06]
Example
W = 101010110111111
W0 = 10101011 W1 = 0111111



ADWIN: A DAPTIVE W ...
The ADWIN Algorithm


Algorithm ADWIN [BG06]
Example
W = 101010110111111 |ˆW0 − µW1 | ≥
                     µ     ˆ      ...
The ADWIN Algorithm


Algorithm ADWIN [BG06]
Example
W = 101010110111111 Drop elements from the tail of W
W0 = 101010110 W...
The ADWIN Algorithm


Algorithm ADWIN [BG06]
Example
W = 01010110111111 Drop elements from the tail of W
W0 = 101010110 W1...
The ADWIN Algorithm


Window Management Models

                                W = 101010110111111

Equal  fixed size subw...
The ADWIN Algorithm


Window Management Models

                                W = 101010110111111

Equal  fixed size subw...
The ADWIN Algorithm


Window Management Models

                                W = 101010110111111

Equal  fixed size subw...
The ADWIN Algorithm


Window Management Models

                                W = 101010110111111

Equal  fixed size subw...
The ADWIN Algorithm


Window Management Models

                                W = 101010110111111

Equal  fixed size subw...
The ADWIN Algorithm


Window Management Models

                                W = 101010110111111

Equal  fixed size subw...
The ADWIN Algorithm


Window Management Models

                                W = 101010110111111

Equal  fixed size subw...
The ADWIN Algorithm


Window Management Models

                                W = 101010110111111

Equal  fixed size subw...
The ADWIN Algorithm


Window Management Models

                                W = 101010110111111

Equal  fixed size subw...
The ADWIN Algorithm


Window Management Models

                                W = 101010110111111

Equal  fixed size subw...
The ADWIN Algorithm


Window Management Models

                                W = 101010110111111

Equal  fixed size subw...
The ADWIN Algorithm


Window Management Models

                                W = 101010110111111

Equal  fixed size subw...
The ADWIN Algorithm


Window Management Models

                                W = 101010110111111

Equal  fixed size subw...
The ADWIN Algorithm


Window Management Models


                                W = 101010110111111

Equal  fixed size sub...
The ADWIN Algorithm


Algorithm ADWIN [BG06]




ADWIN has rigorous guarantees
    On ratio of false positives
    On rati...
The ADWIN Algorithm


Algorithm ADWIN [BG06]

Theorem
At every time step we have:
 1     (Few false positives guarantee) I...
The ADWIN Algorithm


Data Streams Algorithm ADWIN2 [BG06]

ADWIN2 using a Data Stream Sliding Window Model,
    can provi...
The ADWIN Algorithm


Algorithm ADWIN2

ADWIN2 using a Data Stream Sliding Window Model,
    can provide the exact counts ...
The ADWIN Algorithm


Algorithm ADWIN2

ADWIN2 using a Data Stream Sliding Window Model,
    can provide the exact counts ...
The ADWIN Algorithm


Algorithm ADWIN2

ADWIN2 using a Data Stream Sliding Window Model,
    can provide the exact counts ...
The ADWIN Algorithm


Algorithm ADWIN2

ADWIN2 using a Data Stream Sliding Window Model,
    can provide the exact counts ...
The ADWIN Algorithm


Algorithm ADWIN2

ADWIN2 using a Data Stream Sliding Window Model,
    can provide the exact counts ...
The ADWIN Algorithm


Algorithm ADWIN2

ADWIN2 using a Data Stream Sliding Window Model,
    can provide the exact counts ...
The ADWIN Algorithm


Algorithm ADWIN2

ADWIN2 using a Data Stream Sliding Window Model,
    can provide the exact counts ...
The ADWIN Algorithm


Algorithm ADWIN2

ADWIN2 using a Data Stream Sliding Window Model,
    can provide the exact counts ...
General Framework


General Framework Time Change Detectors and
Predictors : Type I

Example (Kalman Filter)
             ...
General Framework


General Framework Time Change Detectors and
Predictors : Type II

Example (Kalman Filter + CUSUM)
    ...
General Framework


General Framework Time Change Detectors and
Predictors : Type III

Example (Adaptive Kalman Filter)
  ...
General Framework


General Framework Time Change Detectors and
Predictors : Type IV

Example (ADWIN, Kalman Filter+ADWIN)...
General Framework


Time Change Detectors and Predictors: A General
Framework


                                    No mem...
General Framework


Time Change Detectors and Predictors: A General
Framework



                               No memory ...
K-ADWIN


K-ADWIN = ADWIN + Kalman Filtering

                                                                            ...
Experimental Validation of K-ADWIN


Tracking Experiments



KALMAN: R=1000;Q=1 Error= 854.97




  A. Bifet, R. Gavaldà (...
Experimental Validation of K-ADWIN


Tracking Experiments
ADWIN : Error= 674.66




KALMAN: R=1000;Q=1 Error= 854.97




 ...
Experimental Validation of K-ADWIN


Tracking Experiments
K-ADWIN Error= 530.13




ADWIN : Error= 674.66




  A. Bifet, ...
Experimental Validation of K-ADWIN


Naïve Bayes

                                   Example
 Data set that               ...
Experimental Validation of K-ADWIN


Naïve Bayes

    Assume we have to classify the following new instance:
     outlook ...
Experimental Validation of K-ADWIN


Experimental Validation of K-ADWIN



    We test Naïve Bayes Predictor and k-means c...
Experimental Validation of K-ADWIN


Naïve Bayes Predictor


                                          Width           %St...
Experimental Validation of K-ADWIN


k -means Clustering

                                                              σ ...
Experimental Validation of K-ADWIN


Results



    No estimator ever does much better than K-ADWIN
    K-ADWIN does much ...
Conclusions


Conclusions and Future Work



    K-ADWIN tunes itself to the data stream at hand, with no need for
    the...
Upcoming SlideShare
Loading in …5
×

Kalman Filters and Adaptive Windows for Learning in Data Streams

1,541 views

Published on

Talk about change detectors, and using Kalman filters with ADWIN, an adaptive sliding window.

Published in: Technology
0 Comments
1 Like
Statistics
Notes
  • Be the first to comment

No Downloads
Views
Total views
1,541
On SlideShare
0
From Embeds
0
Number of Embeds
23
Actions
Shares
0
Downloads
73
Comments
0
Likes
1
Embeds 0
No embeds

No notes for slide

Kalman Filters and Adaptive Windows for Learning in Data Streams

  1. 1. Kalman Filters and Adaptive Windows for Learning in Data Streams Albert Bifet Ricard Gavaldà Universitat Politècnica de Catalunya DS’06 Barcelona A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams Barcelona DS’06 1 / 29
  2. 2. Outline 1 Introduction 2 The Kalman Filter and the CUSUM Test 3 The ADWIN Algorithm 4 General Framework 5 K-ADWIN 6 Experimental Validation of K-ADWIN 7 Conclusions A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams Barcelona DS’06 2 / 29
  3. 3. Introduction Introduction Data Streams Sequence potentially infinite High amount of data: sublinear space High Speed of arrival: small constant time per example Estimation and prediction Distribution and concept drift K-ADWIN : Combination Kalman filter ADWIN : Adaptive window of recently seen data items. A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams Barcelona DS’06 3 / 29
  4. 4. Introduction Introduction Problem Given an input sequence x1 , x2 , . . . , xt , . . . we want to output at instant t a prediction xt+1 minimizing prediction error: |xt+1 − xt+1 | considering distribution changes overtime. A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams Barcelona DS’06 4 / 29
  5. 5. Introduction Time Change Detectors and Predictors: A General Framework Estimation - xt - Estimator A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams Barcelona DS’06 5 / 29
  6. 6. Introduction Time Change Detectors and Predictors: A General Framework Estimation - xt - Estimator Alarm - - Change Detect. A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams Barcelona DS’06 5 / 29
  7. 7. Introduction Time Change Detectors and Predictors: A General Framework Estimation - xt - Estimator Alarm - - Change Detect. 6 6 ? - Memory A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams Barcelona DS’06 5 / 29
  8. 8. Introduction Introduction Our particular proposal: K-ADWIN Our generic proposal: Kalman filter as estimator Use change detector Use ADWIN as change Use memory detector with memory [BG06] Application Estimate statistics from data streams In Data Mining Algorithms based on counters, replace them for estimators. A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams Barcelona DS’06 6 / 29
  9. 9. Introduction Data Mining Algorithms with Concept Drift No Concept Drift Concept drift DM Algorithm DM Algorithm input output input output - Counter5 - - - Counter4 Static Model Counter3 6 Counter2 Counter1 - Change Detect. A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams Barcelona DS’06 7 / 29
  10. 10. Introduction Data Mining Algorithms with Concept Drift No Concept Drift Concept Drift DM Algorithm DM Algorithm input output input output - Counter5 - - Estimator5 - Counter4 Estimator4 Counter3 Estimator3 Counter2 Estimator2 Counter1 Estimator1 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams Barcelona DS’06 7 / 29
  11. 11. The Kalman Filter and the CUSUM Test The Kalman Filter Optimal recursive algorithm Minimum mean-square error estimator Estimate the state x ∈ n of a discrete-time controlled process xk = Axk −1 + Buk + wk −1 with a measurement z ∈ m that is Zk = Hxk + vk . The random variables wk and vk represent the process and measurement noise (respectively). They are assumed to be independent (of each other), white, and with normal probability distributions p(w) ∼ N(0, Q) p(v ) ∼ N(0, R). A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams Barcelona DS’06 8 / 29
  12. 12. The Kalman Filter and the CUSUM Test The Kalman Filter The difference equation of our discrete-time controlled process is Kk = Pk −1 /(Pk −1 + R) Xk = Xk −1 + Kk (zk − Xk −1 ) Pk = Pk (1 − Kk ) + Q A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams Barcelona DS’06 9 / 29
  13. 13. The Kalman Filter and the CUSUM Test The Kalman Filter The difference equation of our discrete-time controlled process is Kk = Pk −1 /(Pk −1 + R) Xk = Xk −1 + Kk (zk − Xk −1 ) Pk = Pk (1 − Kk ) + Q The performance of the Kalman filter depends on the accuracy of the a-priori assumptions: linearity of the difference stochastic equation estimation of covariances Q and R, assumed to be fixed, known, and follow normal distributions with zero mean. A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams Barcelona DS’06 9 / 29
  14. 14. The Kalman Filter and the CUSUM Test The CUSUM Test The cumulative sum (CUSUM algorithm),is a change detection algorithm that gives an alarm when the mean of the input data is significantly different from zero. The CUSUM test is memoryless, and its accuracy depends on the choice of parameters υ and h. It is as follows: g0 = 0, gt = max (0, gt−1 + t − υ) if gt h then alarm and gt = 0 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 10 / 29
  15. 15. The ADWIN Algorithm Algorithm ADWIN [BG06] Example W = 101010110111111 ADWIN: A DAPTIVE W INDOWING A LGORITHM 1 Initialize Window W 2 for each t 0 3 do W ← W ∪ {xt } (i.e., add xt to the head of W ) 4 repeat Drop elements from the tail of W 5 until |ˆW0 − µW1 | ≥ c holds µ ˆ 6 for every split of W into W = W0 · W1 7 Output µW ˆ A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 11 / 29
  16. 16. The ADWIN Algorithm Algorithm ADWIN [BG06] Example W = 101010110111111 W0 = 1 W1 = 01010110111111 ADWIN: A DAPTIVE W INDOWING A LGORITHM 1 Initialize Window W 2 for each t 0 3 do W ← W ∪ {xt } (i.e., add xt to the head of W ) 4 repeat Drop elements from the tail of W 5 until |ˆW0 − µW1 | ≥ c holds µ ˆ 6 for every split of W into W = W0 · W1 7 Output µW ˆ A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 11 / 29
  17. 17. The ADWIN Algorithm Algorithm ADWIN [BG06] Example W = 101010110111111 W0 = 10 W1 = 1010110111111 ADWIN: A DAPTIVE W INDOWING A LGORITHM 1 Initialize Window W 2 for each t 0 3 do W ← W ∪ {xt } (i.e., add xt to the head of W ) 4 repeat Drop elements from the tail of W 5 until |ˆW0 − µW1 | ≥ c holds µ ˆ 6 for every split of W into W = W0 · W1 7 Output µW ˆ A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 11 / 29
  18. 18. The ADWIN Algorithm Algorithm ADWIN [BG06] Example W = 101010110111111 W0 = 101 W1 = 010110111111 ADWIN: A DAPTIVE W INDOWING A LGORITHM 1 Initialize Window W 2 for each t 0 3 do W ← W ∪ {xt } (i.e., add xt to the head of W ) 4 repeat Drop elements from the tail of W 5 until |ˆW0 − µW1 | ≥ c holds µ ˆ 6 for every split of W into W = W0 · W1 7 Output µW ˆ A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 11 / 29
  19. 19. The ADWIN Algorithm Algorithm ADWIN [BG06] Example W = 101010110111111 W0 = 1010 W1 = 10110111111 ADWIN: A DAPTIVE W INDOWING A LGORITHM 1 Initialize Window W 2 for each t 0 3 do W ← W ∪ {xt } (i.e., add xt to the head of W ) 4 repeat Drop elements from the tail of W 5 until |ˆW0 − µW1 | ≥ c holds µ ˆ 6 for every split of W into W = W0 · W1 7 Output µW ˆ A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 11 / 29
  20. 20. The ADWIN Algorithm Algorithm ADWIN [BG06] Example W = 101010110111111 W0 = 10101 W1 = 0110111111 ADWIN: A DAPTIVE W INDOWING A LGORITHM 1 Initialize Window W 2 for each t 0 3 do W ← W ∪ {xt } (i.e., add xt to the head of W ) 4 repeat Drop elements from the tail of W 5 until |ˆW0 − µW1 | ≥ c holds µ ˆ 6 for every split of W into W = W0 · W1 7 Output µW ˆ A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 11 / 29
  21. 21. The ADWIN Algorithm Algorithm ADWIN [BG06] Example W = 101010110111111 W0 = 101010 W1 = 110111111 ADWIN: A DAPTIVE W INDOWING A LGORITHM 1 Initialize Window W 2 for each t 0 3 do W ← W ∪ {xt } (i.e., add xt to the head of W ) 4 repeat Drop elements from the tail of W 5 until |ˆW0 − µW1 | ≥ c holds µ ˆ 6 for every split of W into W = W0 · W1 7 Output µW ˆ A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 11 / 29
  22. 22. The ADWIN Algorithm Algorithm ADWIN [BG06] Example W = 101010110111111 W0 = 1010101 W1 = 10111111 ADWIN: A DAPTIVE W INDOWING A LGORITHM 1 Initialize Window W 2 for each t 0 3 do W ← W ∪ {xt } (i.e., add xt to the head of W ) 4 repeat Drop elements from the tail of W 5 until |ˆW0 − µW1 | ≥ c holds µ ˆ 6 for every split of W into W = W0 · W1 7 Output µW ˆ A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 11 / 29
  23. 23. The ADWIN Algorithm Algorithm ADWIN [BG06] Example W = 101010110111111 W0 = 10101011 W1 = 0111111 ADWIN: A DAPTIVE W INDOWING A LGORITHM 1 Initialize Window W 2 for each t 0 3 do W ← W ∪ {xt } (i.e., add xt to the head of W ) 4 repeat Drop elements from the tail of W 5 until |ˆW0 − µW1 | ≥ c holds µ ˆ 6 for every split of W into W = W0 · W1 7 Output µW ˆ A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 11 / 29
  24. 24. The ADWIN Algorithm Algorithm ADWIN [BG06] Example W = 101010110111111 |ˆW0 − µW1 | ≥ µ ˆ c : CHANGE DETECTED! W0 = 101010110 W1 = 111111 ADWIN: A DAPTIVE W INDOWING A LGORITHM 1 Initialize Window W 2 for each t 0 3 do W ← W ∪ {xt } (i.e., add xt to the head of W ) 4 repeat Drop elements from the tail of W 5 until |ˆW0 − µW1 | ≥ c holds µ ˆ 6 for every split of W into W = W0 · W1 7 Output µW ˆ A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 11 / 29
  25. 25. The ADWIN Algorithm Algorithm ADWIN [BG06] Example W = 101010110111111 Drop elements from the tail of W W0 = 101010110 W1 = 111111 ADWIN: A DAPTIVE W INDOWING A LGORITHM 1 Initialize Window W 2 for each t 0 3 do W ← W ∪ {xt } (i.e., add xt to the head of W ) 4 repeat Drop elements from the tail of W 5 until |ˆW0 − µW1 | ≥ c holds µ ˆ 6 for every split of W into W = W0 · W1 7 Output µW ˆ A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 11 / 29
  26. 26. The ADWIN Algorithm Algorithm ADWIN [BG06] Example W = 01010110111111 Drop elements from the tail of W W0 = 101010110 W1 = 111111 ADWIN: A DAPTIVE W INDOWING A LGORITHM 1 Initialize Window W 2 for each t 0 3 do W ← W ∪ {xt } (i.e., add xt to the head of W ) 4 repeat Drop elements from the tail of W 5 until |ˆW0 − µW1 | ≥ c holds µ ˆ 6 for every split of W into W = W0 · W1 7 Output µW ˆ A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 11 / 29
  27. 27. The ADWIN Algorithm Window Management Models W = 101010110111111 Equal fixed size subwindows Total window against subwindow 1010 1011011 1111 10101011011 1111 D. Kifer, S. Ben-David, and J. J. Gama, P. Medas, G. Castillo, Gehrke. Detecting change in data and P. Rodrigues. Learning with streams. 2004 drift detection. 2004 ADWIN: All Adjacent subwindows 1 01010110111111 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 12 / 29
  28. 28. The ADWIN Algorithm Window Management Models W = 101010110111111 Equal fixed size subwindows Total window against subwindow 1010 1011011 1111 10101011011 1111 D. Kifer, S. Ben-David, and J. J. Gama, P. Medas, G. Castillo, Gehrke. Detecting change in data and P. Rodrigues. Learning with streams. 2004 drift detection. 2004 ADWIN: All Adjacent subwindows 10 1010110111111 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 12 / 29
  29. 29. The ADWIN Algorithm Window Management Models W = 101010110111111 Equal fixed size subwindows Total window against subwindow 1010 1011011 1111 10101011011 1111 D. Kifer, S. Ben-David, and J. J. Gama, P. Medas, G. Castillo, Gehrke. Detecting change in data and P. Rodrigues. Learning with streams. 2004 drift detection. 2004 ADWIN: All Adjacent subwindows 101 010110111111 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 12 / 29
  30. 30. The ADWIN Algorithm Window Management Models W = 101010110111111 Equal fixed size subwindows Total window against subwindow 1010 1011011 1111 10101011011 1111 D. Kifer, S. Ben-David, and J. J. Gama, P. Medas, G. Castillo, Gehrke. Detecting change in data and P. Rodrigues. Learning with streams. 2004 drift detection. 2004 ADWIN: All Adjacent subwindows 1010 10110111111 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 12 / 29
  31. 31. The ADWIN Algorithm Window Management Models W = 101010110111111 Equal fixed size subwindows Total window against subwindow 1010 1011011 1111 10101011011 1111 D. Kifer, S. Ben-David, and J. J. Gama, P. Medas, G. Castillo, Gehrke. Detecting change in data and P. Rodrigues. Learning with streams. 2004 drift detection. 2004 ADWIN: All Adjacent subwindows 10101 0110111111 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 12 / 29
  32. 32. The ADWIN Algorithm Window Management Models W = 101010110111111 Equal fixed size subwindows Total window against subwindow 1010 1011011 1111 10101011011 1111 D. Kifer, S. Ben-David, and J. J. Gama, P. Medas, G. Castillo, Gehrke. Detecting change in data and P. Rodrigues. Learning with streams. 2004 drift detection. 2004 ADWIN: All Adjacent subwindows 101010 110111111 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 12 / 29
  33. 33. The ADWIN Algorithm Window Management Models W = 101010110111111 Equal fixed size subwindows Total window against subwindow 1010 1011011 1111 10101011011 1111 D. Kifer, S. Ben-David, and J. J. Gama, P. Medas, G. Castillo, Gehrke. Detecting change in data and P. Rodrigues. Learning with streams. 2004 drift detection. 2004 ADWIN: All Adjacent subwindows 1010101 10111111 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 12 / 29
  34. 34. The ADWIN Algorithm Window Management Models W = 101010110111111 Equal fixed size subwindows Total window against subwindow 1010 1011011 1111 10101011011 1111 D. Kifer, S. Ben-David, and J. J. Gama, P. Medas, G. Castillo, Gehrke. Detecting change in data and P. Rodrigues. Learning with streams. 2004 drift detection. 2004 ADWIN: All Adjacent subwindows 10101011 0111111 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 12 / 29
  35. 35. The ADWIN Algorithm Window Management Models W = 101010110111111 Equal fixed size subwindows Total window against subwindow 1010 1011011 1111 10101011011 1111 D. Kifer, S. Ben-David, and J. J. Gama, P. Medas, G. Castillo, Gehrke. Detecting change in data and P. Rodrigues. Learning with streams. 2004 drift detection. 2004 ADWIN: All Adjacent subwindows 101010110 111111 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 12 / 29
  36. 36. The ADWIN Algorithm Window Management Models W = 101010110111111 Equal fixed size subwindows Total window against subwindow 1010 1011011 1111 10101011011 1111 D. Kifer, S. Ben-David, and J. J. Gama, P. Medas, G. Castillo, Gehrke. Detecting change in data and P. Rodrigues. Learning with streams. 2004 drift detection. 2004 ADWIN: All Adjacent subwindows 1010101101 11111 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 12 / 29
  37. 37. The ADWIN Algorithm Window Management Models W = 101010110111111 Equal fixed size subwindows Total window against subwindow 1010 1011011 1111 10101011011 1111 D. Kifer, S. Ben-David, and J. J. Gama, P. Medas, G. Castillo, Gehrke. Detecting change in data and P. Rodrigues. Learning with streams. 2004 drift detection. 2004 ADWIN: All Adjacent subwindows 10101011011 1111 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 12 / 29
  38. 38. The ADWIN Algorithm Window Management Models W = 101010110111111 Equal fixed size subwindows Total window against subwindow 1010 1011011 1111 10101011011 1111 D. Kifer, S. Ben-David, and J. J. Gama, P. Medas, G. Castillo, Gehrke. Detecting change in data and P. Rodrigues. Learning with streams. 2004 drift detection. 2004 ADWIN: All Adjacent subwindows 101010110111 111 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 12 / 29
  39. 39. The ADWIN Algorithm Window Management Models W = 101010110111111 Equal fixed size subwindows Total window against subwindow 1010 1011011 1111 10101011011 1111 D. Kifer, S. Ben-David, and J. J. Gama, P. Medas, G. Castillo, Gehrke. Detecting change in data and P. Rodrigues. Learning with streams. 2004 drift detection. 2004 ADWIN: All Adjacent subwindows 1010101101111 11 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 12 / 29
  40. 40. The ADWIN Algorithm Window Management Models W = 101010110111111 Equal fixed size subwindows Total window against subwindow 1010 1011011 1111 10101011011 1111 D. Kifer, S. Ben-David, and J. J. Gama, P. Medas, G. Castillo, Gehrke. Detecting change in data and P. Rodrigues. Learning with streams. 2004 drift detection. 2004 ADWIN: All Adjacent subwindows 10101011011111 1 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 12 / 29
  41. 41. The ADWIN Algorithm Algorithm ADWIN [BG06] ADWIN has rigorous guarantees On ratio of false positives On ratio of false negatives On the relation of the size of the current window and change rates A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 13 / 29
  42. 42. The ADWIN Algorithm Algorithm ADWIN [BG06] Theorem At every time step we have: 1 (Few false positives guarantee) If µt remains constant within W , the probability that ADWIN shrinks the window at this step is at most δ. 2 (Few false negatives guarantee) If for any partition W in two parts W0 W1 (where W1 contains the most recent items) we have |µW0 − µW1 | , and if 3 max{µW0 , µW1 } 4n ≥4· ln min{n0 , n1 } δ then with probability 1 − δ ADWIN shrinks W to W1 , or shorter. A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 14 / 29
  43. 43. The ADWIN Algorithm Data Streams Algorithm ADWIN2 [BG06] ADWIN2 using a Data Stream Sliding Window Model, can provide the exact counts of 1’s in O(1) time per point. tries O(log W ) cutpoints uses O( 1 log W ) memory words the processing time per example is O(log W ) (amortized) and O(log2 W ) (worst-case). Sliding Window Model 1010101 101 11 1 1 Content: 4 2 2 1 1 Capacity: 7 3 2 1 1 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 15 / 29
  44. 44. The ADWIN Algorithm Algorithm ADWIN2 ADWIN2 using a Data Stream Sliding Window Model, can provide the exact counts of 1’s in O(1) time per point. tries O(log W ) cutpoints uses O( 1 log W ) memory words the processing time per example is O(log W ) (amortized) and O(log2 W ) (worst-case). Insert new Item 1010101 101 11 1 1 Content: 4 2 2 1 1 Capacity: 7 3 2 1 1 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 16 / 29
  45. 45. The ADWIN Algorithm Algorithm ADWIN2 ADWIN2 using a Data Stream Sliding Window Model, can provide the exact counts of 1’s in O(1) time per point. tries O(log W ) cutpoints uses O( 1 log W ) memory words the processing time per example is O(log W ) (amortized) and O(log2 W ) (worst-case). Insert new Item 1010101 101 11 1 1 1 Content: 4 2 2 1 1 1 Capacity: 7 3 2 1 1 1 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 16 / 29
  46. 46. The ADWIN Algorithm Algorithm ADWIN2 ADWIN2 using a Data Stream Sliding Window Model, can provide the exact counts of 1’s in O(1) time per point. tries O(log W ) cutpoints uses O( 1 log W ) memory words the processing time per example is O(log W ) (amortized) and O(log2 W ) (worst-case). Compressing Buckets 1010101 101 11 1 1 1 Content: 4 2 2 1 1 1 Capacity: 7 3 2 1 1 1 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 16 / 29
  47. 47. The ADWIN Algorithm Algorithm ADWIN2 ADWIN2 using a Data Stream Sliding Window Model, can provide the exact counts of 1’s in O(1) time per point. tries O(log W ) cutpoints uses O( 1 log W ) memory words the processing time per example is O(log W ) (amortized) and O(log2 W ) (worst-case). Compressing Buckets 1010101 101 11 11 1 Content: 4 2 2 2 1 Capacity: 7 3 2 2 1 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 17 / 29
  48. 48. The ADWIN Algorithm Algorithm ADWIN2 ADWIN2 using a Data Stream Sliding Window Model, can provide the exact counts of 1’s in O(1) time per point. tries O(log W ) cutpoints uses O( 1 log W ) memory words the processing time per example is O(log W ) (amortized) and O(log2 W ) (worst-case). Compressing Buckets 1010101 101 11 11 1 Content: 4 2 2 2 1 Capacity: 7 3 2 2 1 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 17 / 29
  49. 49. The ADWIN Algorithm Algorithm ADWIN2 ADWIN2 using a Data Stream Sliding Window Model, can provide the exact counts of 1’s in O(1) time per point. tries O(log W ) cutpoints uses O( 1 log W ) memory words the processing time per example is O(log W ) (amortized) and O(log2 W ) (worst-case). Compressing Buckets 1010101 10111 11 1 Content: 4 4 2 1 Capacity: 7 5 2 1 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 18 / 29
  50. 50. The ADWIN Algorithm Algorithm ADWIN2 ADWIN2 using a Data Stream Sliding Window Model, can provide the exact counts of 1’s in O(1) time per point. tries O(log W ) cutpoints uses O( 1 log W ) memory words the processing time per example is O(log W ) (amortized) and O(log2 W ) (worst-case). Detecting Change: Delete last Bucket 1010101 10111 11 1 Content: 4 4 2 1 Capacity: 7 5 2 1 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 19 / 29
  51. 51. The ADWIN Algorithm Algorithm ADWIN2 ADWIN2 using a Data Stream Sliding Window Model, can provide the exact counts of 1’s in O(1) time per point. tries O(log W ) cutpoints uses O( 1 log W ) memory words the processing time per example is O(log W ) (amortized) and O(log2 W ) (worst-case). Detecting Change: Delete last Bucket 10111 11 1 Content: 4 2 1 Capacity: 5 2 1 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 19 / 29
  52. 52. General Framework General Framework Time Change Detectors and Predictors : Type I Example (Kalman Filter) Estimation - xt - Estimator A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 20 / 29
  53. 53. General Framework General Framework Time Change Detectors and Predictors : Type II Example (Kalman Filter + CUSUM) Estimation - xt - Estimator Alarm - - Change Detect. A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 20 / 29
  54. 54. General Framework General Framework Time Change Detectors and Predictors : Type III Example (Adaptive Kalman Filter) Estimation - xt - Estimator 6 - Memory A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 20 / 29
  55. 55. General Framework General Framework Time Change Detectors and Predictors : Type IV Example (ADWIN, Kalman Filter+ADWIN) Estimation - xt - Estimator Alarm - - Change Detect. 6 6 ? - Memory A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 20 / 29
  56. 56. General Framework Time Change Detectors and Predictors: A General Framework No memory Memory No Change Type I Type III Detector Kalman Filter Adaptive Kalman Filter Change Type II Type IV Detector Kalman Filter + CUSUM ADWIN Kalman Filter + ADWIN A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 21 / 29
  57. 57. General Framework Time Change Detectors and Predictors: A General Framework No memory Memory No Change Type I Type III Detector Kalman Filter Adaptive Kalman Filter Q,R estimated from window Change Type II Type IV Detector Kalman Filter + CUSUM ADWIN Kalman Filter + ADWIN Q,R estimated from window A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 21 / 29
  58. 58. K-ADWIN K-ADWIN = ADWIN + Kalman Filtering Estimation - xt - Kalman Alarm - - ADWIN 6 6 ? - ADWIN Memory R = W 2 /50 and Q = 200/W , where W is the length of the window maintained by ADWIN. A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 22 / 29
  59. 59. Experimental Validation of K-ADWIN Tracking Experiments KALMAN: R=1000;Q=1 Error= 854.97 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 23 / 29
  60. 60. Experimental Validation of K-ADWIN Tracking Experiments ADWIN : Error= 674.66 KALMAN: R=1000;Q=1 Error= 854.97 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 23 / 29
  61. 61. Experimental Validation of K-ADWIN Tracking Experiments K-ADWIN Error= 530.13 ADWIN : Error= 674.66 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 23 / 29
  62. 62. Experimental Validation of K-ADWIN Naïve Bayes Example Data set that outlook temp. humidity windy play describes the sunny hot high false no weather sunny hot high true no conditions for overcast hot high false yes playing some rainy mild high false yes game. rainy cool normal false yes rainy cool normal true no overcast cool normal true yes Assume we have to classify the following new instance: outlook temp. humidity windy play sunny cool high true ? A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 24 / 29
  63. 63. Experimental Validation of K-ADWIN Naïve Bayes Assume we have to classify the following new instance: outlook temp. humidity windy play sunny cool high true ? We classify the new instance: νNB = arg max P(νj )P(sunny |νj )P(cool|νj )P(high|νj )P(true|νj ) ν∈{yes,no} Conditional probabilities can be estimated directly as frequencies: number of instances with attribute ai and class νj P(ai |νj ) = total number of training instances with class νj Create one estimator for each frequence that needs estimation A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 24 / 29
  64. 64. Experimental Validation of K-ADWIN Experimental Validation of K-ADWIN We test Naïve Bayes Predictor and k-means clustering Method: replace counters by estimators Synthetic data where change is controllable Naïve Bayes: We compare accuracy of Static model: Training of 1000 samples every instant Dynamic model: replace probabilities counters by estimators %Dynamic computing the ratio Static with tests using 2000 samples. A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 25 / 29
  65. 65. Experimental Validation of K-ADWIN Naïve Bayes Predictor Width %Static %Dynamic % Dynamic/Static ADWIN 83,36% 80,30% 96,33% Kalman Q = 1, R = 1000 83,22% 71,13% 85,48% Kalman Q = 1, R = 1 83,21% 56,91% 68,39% Kalman Q = .25, R = .25 83,26% 56,91% 68,35% Adaptive Kalman 83,24% 76,21% 91,56% CUSUM Kalman 83,30% 50,65% 60,81% K-ADWIN 83,24% 81,39% 97,77% Fixed-sized Window 32 83,28% 67,64% 81,22% Fixed-sized Window 128 83,30% 75,40% 90,52% Fixed-sized Window 512 83,28% 80,47% 96,62% Fixed-sized Window 2048 83,24% 82,19% 98,73% A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 26 / 29
  66. 66. Experimental Validation of K-ADWIN k -means Clustering σ = 0.15 Width Static Dynamic ADWIN 9,72 21,54 Kalman Q = 1, R = 1000 9,72 19,72 Kalman Q = 1, R = 100 9,71 17,60 Kalman Q = .25, R = .25 9,71 22,63 Adaptive Kalman 9,72 18,98 CUSUM Kalman 9,72 18,29 K-ADWIN 9,72 17,30 Fixed-sized Window 32 9,72 25,70 Fixed-sized Window 128 9,72 36,42 Fixed-sized Window 512 9,72 38,75 Fixed-sized Window 2048 9,72 39,64 Fixed-sized Window 8192 9,72 43,39 Fixed-sized Window 32768 9,72 53,82 A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 27 / 29
  67. 67. Experimental Validation of K-ADWIN Results No estimator ever does much better than K-ADWIN K-ADWIN does much better than every other estimators in at least one context. Tracking problem K-ADWIN and ADWIN automatically do about as well as the Kalman filter with the best set of fixed covariance parameters. Naïve Bayes and k -means: K-ADWIN does somewhat better than ADWIN and far better than any memoryless Kalman filter. A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 28 / 29
  68. 68. Conclusions Conclusions and Future Work K-ADWIN tunes itself to the data stream at hand, with no need for the user to hardwire or precompute parameters. Better results than either memoryless Kalman Filtering or sliding windows with linear estimators. Future work : Tests on real-world, not only synthetic data. Other learning algorithms: algorithms for induction of decision trees. A. Bifet, R. Gavaldà (UPC) Kalman Filters and Adaptive Windows for Learning in Data Streams DS’06 Barcelona 29 / 29

×