Real-Time Machine Learning with Node.js - Philipp Burckhardt, Carnegie Mellon University

11/14/2016 Machine Learning
http://localhost:3000/#/?export&_k=lv9ﬂd 1/23

REAL-TIME MACHINE
LEARNING WITH NODE.JS
PHILIPP BURCKHARDT
Carnegie Mellon University

LEARNING
PATTERNS
FROM DATA(iStock)

REAL-TIME MACHINE
LEARNING WITH
NODE.JS

BATCH
Build model using a batch of available data
INCREMENTAL
Update model as new data comes in
TRAINING ALGORITHMS

8. // For each simulated datum, update the
mean...
9. for ( var i = 0; i < 100; i++ ) {
10. var v = randu() * 100.0;
11. accumulator( v );
12. }
13. var mean = accumulator();
14.
15.
4. var incrmean = require(
'@stdlib/math/generics/statistics/incrmean' );
5.
6. var accumulator = incrmean();
7.
16.
17.
Update estimator as new data comes in...

Prediction is very di cult,
especially if it's about the
future.
- Nils Bohr

INDEPENDENTLY AND
IDENTICALLY
DISTRIBUTED (I.I.D.)
DATA ASSUMED TO BE
Might not hold: e.g., time series are mostly non-stationary

8. setInterval( function() {
9. var mem = os.freemem() / os.totalmem();
10. accumulator( mem );
11. var mean = accumulator();
12. }, 1000 );
13.
14.
15.
16.
17.
1. 'use strict';
2.
3. var incrmmean = require(
'@stdlib/math/generics/statistics/incrmmean' );
4. var os = require( 'os' );
5.
6. var accumulator = incrmmean( 5 );
7.
Update moving mean as data comes in...

Moving Means
window size

TYPES OF PROBLEMS
Regression
0 20 40 60 80 100
1,000
500
0
500
1,000
1,500
e.g., house prices
Classi cation
0 20 40 60 80 100
0
20
40
60
80
100
e.g., character recognition (OCR)
Clustering
0 20 40 60 80 100
0
20
40
60
80
100
e.g., movie tastes

REGRESSION
Model relationship between a numeric dependent
variable y and one or more explanatory variables X.

14. registry
15. .on( 'package', function onPkg( pkg ) {
16. var nVersions = pkg.versions ?
17. pkg.versions.length : 0;
18. if ( pkg.created ) {
19. var current = new Date().getTime();
20. var created = new Date( pkg.created );
21. var age = ( current - created ) /
22. ( 1000 * 60 * 60 * 24 * 365 );
23. model.update( [ age ], nVersions );
24. }
25. })
10. 'loss': 'huber',
11. 'intercept': true
12. });
13.
26.
27.
28.
29.
30.
31.
Use creation date to predict # of versions

Start Regression line: = 0.000 + 0.000x
Number of package versions is positively correlated with age:
y^

CLASSIFICATION
Model relationship between a dependent categorical
variable y and one or more explanatory variables X.

Predicting a binary outcome
8. var model = onlineClassification({
9. 'lambda': 1e-6,
10. 'intercept': true,
11. 'loss': 'log'
12. });
13.
14. registry.on( 'package', function onPkg( pkg )
{
15. var usesReact = pkg.mentions( 'react' ) ?
16. 1 :
17. -1;
18.
19. var features = [
20. 'webpack', 'browserify', 'jest',
21. 'tape', 'mocha'
22. ].map(
23. d => pkg.devDependsOn( d ) );
24.
25. var phat = model.predict( features,
'probability' );
26. var yhat = phat > 0.5 ? +1 : -1;
27.

React Usage
paradigm-tags
marketeer
couchcache
datamodel-to-openapi
eaze-request
paradigm-categories
joeljparks-hubot-cosmicjr
paradigm-taxonomies
gulp-controlled-merge-json
ember-cli-addon-tests
Predicted
Yes
Predicted
No
Yes 0 4
No 0 34
Webpack Browserify Jest Tape Mocha
2.0
1.0
1.0
2.0

Evaluating regression and classi cation
models
500 1,000 1,500 2,000 2,500
0.10
0.20
0.30
0.40
0.50
15%
Look at generalization
error (performance on data not
used for model training)
Our toy model does not do
so well: A mis-classi cation rate
of 13% might sound great, but
always predicting -1 yields 15%!

CLUSTERING
Group observations into meaningful clusters such that
objects within a cluster are similar to each other and di erent
from objects assigned to the other clusters.

POPULAR ALGORITHMS
kmeans
dbscan
Hierarchical Clustering
Mixture of Gaussians
4
5
6
7
8Dim. 1
2
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
3
3.1
3.2
3.3
3.4
3.5
3.6
3.7
3.8
3.9
4
4.1
4.2
4.3
4.4
Dim. 2
1
2
3
4
5
6
7
Dim. 3
4
5
6
7
8 Dim. 1
2
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
3
3.1
3.2
3.3
3.4
3.5
3.6
3.7
3.8
3.9
4
4.1
4.2
4.3
4.4
Dim. 2
1
2
3
4
5
6
7
Dim. 3
Cluster 1
Cluster 2
Cluster 3
Iris setosa
Iris versicolor
Iris virginica
Iris SpecieskMeans Clusters

Free Textbooks:
"An Introduction to Statistical Learning" by James, Witten,
Hastie & Tibshirani (plus accompanying video lectures)
"Elements of Statistical Learning: Data Mining, Inference,
and Prediction." by Hastie, Tibshirani & Friedman
stdlib GitHub repository: https://github.com/stdlib-
js/stdlib/tree/develop
FURTHER RESOURCES

THANK YOU!

Real-Time Machine Learning with Node.js - Philipp Burckhardt, Carnegie Mellon University

More Related Content

Viewers also liked (19)

Similar to Real-Time Machine Learning with Node.js - Philipp Burckhardt, Carnegie Mellon University (20)

More from NodejsFoundation (6)

Recently uploaded (20)

Real-Time Machine Learning with Node.js - Philipp Burckhardt, Carnegie Mellon University