Big Data Day LA 2016/ Hadoop/ Spark/ Kafka track - Real-time Aggregations, Approxs, Similarities, and Recommendations at Scale using Spark Kafka, Docker, CoreNLP, Word2Vec, LDA, and Twitter Algebird - Chris Fregly, Research Scientist, PipelineIO
5 likes647 views
Live, Interactive Recommendations Demo Spark Streaming, ML, GraphX, Kafka, Cassandra, Docker, CoreNLP, Word2Vec, LDA, and Twitter Algebird (advancedspark.com). Types of Similarity - Euclidean vs. Non-Euclidean, Similarity, Jaccard Similarity, Cosine Similarity, LogLikelihood Similarity, Edit Distance. Text-based Similarities and Analytics - Word2Vec, LDA Topic Extraction, TextRank, Similarity-based Recommendations, User-to-User, Content-based, Item-to-Item (Amazon), Collaborative-based, User-to-Item (Netflix), Graph-based, Item-to-Item ""Pathways"" (Spotify). Aggregations, Approximations, and Similarities at Scale - Twitter Algebird, MinHash and Bucketing, Locality Sensitive Hashing (LSH), BloomFilters, CountMin Sketch, HyperLogLog
![Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production
Feature Engineering
Dimension Reduction
Reduce number of features in feature space
Principle Component Analysis (PCA)
Find principle features that best describe data variance
Peel dimensional layers back
One-Hot Encoding
Convert nominal categorical feature values into 0’s and 1’s
Remove any numerical relationship between categories
Bears -> 1 Bears -> [1.0, 0.0, 0.0]
49’ers -> 2 --> 49’ers -> [0.0, 1.0, 0.0]
Steelers-> 3 Steelers-> [0.0, 0.0, 1.0]
35
Convert Each Item
to Binary Vector
with Single 1.0 Column](https://image.slidesharecdn.com/bigdatadaylajuly2016-160812175025/85/Big-Data-Day-LA-2016-Hadoop-Spark-Kafka-track-Real-time-Aggregations-Approxs-Similarities-and-Recommendations-at-Scale-using-Spark-Kafka-Docker-CoreNLP-Word2Vec-LDA-and-Twitter-Algebird-Chris-Fregly-Research-Scientist-PipelineIO-35-320.jpg)
![Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production
Feature Normalization & Standardization
Goal
Scale features to standard size
Prevent boundless features
Helps avoid overfitting
Required by many ML algos
Normalize Features
Calculate L1 (or L2, etc) norm, then divide into each element
Standardize Features
Apply standard normal transformation (mean->0, stddev->1)
org.apache.spark.ml.feature.[Normalizer, StandardScaler]
36
http://www.mathsisfun.com/data/standard-normal-distribution.html](https://image.slidesharecdn.com/bigdatadaylajuly2016-160812175025/85/Big-Data-Day-LA-2016-Hadoop-Spark-Kafka-track-Real-time-Aggregations-Approxs-Similarities-and-Recommendations-at-Scale-using-Spark-Kafka-Docker-CoreNLP-Word2Vec-LDA-and-Twitter-Algebird-Chris-Fregly-Research-Scientist-PipelineIO-36-320.jpg)
![Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production
Item-to-Item Collaborative Filtering
Famous Amazon Paper circa 2003
Problem
As users grew, user-to-item collaborative filtering didn’t scale
Solution
Item-to-item similarity, nearest neighbors
Offline (Batch)
Generate itemId->List[userId] vectors
Online (Real-time)
From cart, recommend nearest-neighbors in vector space
58](https://image.slidesharecdn.com/bigdatadaylajuly2016-160812175025/85/Big-Data-Day-LA-2016-Hadoop-Spark-Kafka-track-Real-time-Aggregations-Approxs-Similarities-and-Recommendations-at-Scale-using-Spark-Kafka-Docker-CoreNLP-Word2Vec-LDA-and-Twitter-Algebird-Chris-Fregly-Research-Scientist-PipelineIO-58-320.jpg)
![Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production
Approximations in Spark Libraries
Spark Core
countByKeyApprox(timeout: Long, confidence: Double)
PartialResult
Spark SQL
approxCountDistinct(column: Column, targetResidual: Float)
approxQuantile(column: Column, quantiles: Seq[Float], targetResidual: Float)
Spark ML
Stratified sampling
sampleByKey(fractions: Map[K, Double])
DIMSUM sampling
Probabilistic sampling reduces amount of shuffle
RowMatrix.columnSimilarities(threshold: Double)
92](https://image.slidesharecdn.com/bigdatadaylajuly2016-160812175025/85/Big-Data-Day-LA-2016-Hadoop-Spark-Kafka-track-Real-time-Aggregations-Approxs-Similarities-and-Recommendations-at-Scale-using-Spark-Kafka-Docker-CoreNLP-Word2Vec-LDA-and-Twitter-Algebird-Chris-Fregly-Research-Scientist-PipelineIO-92-320.jpg)
![[QE 2018] Łukasz Gawron – Testing Batch and Streaming Spark Applications](https://cdn.slidesharecdn.com/ss_thumbnails/gawrontestingbatchandstreamingsparkapplications-180702090002-thumbnail.jpg?width=560&fit=bounds)
![Scott Anderson [InfluxData] | InfluxDB Tasks – Beyond Downsampling | InfluxDa...](https://cdn.slidesharecdn.com/ss_thumbnails/influxdaysnov20201-201105232414-thumbnail.jpg?width=560&fit=bounds)
Ad
More Related Content
What's hot (20)
Viewers also liked (20)
Ad
Similar to Big Data Day LA 2016/ Hadoop/ Spark/ Kafka track - Real-time Aggregations, Approxs, Similarities, and Recommendations at Scale using Spark Kafka, Docker, CoreNLP, Word2Vec, LDA, and Twitter Algebird - Chris Fregly, Research Scientist, PipelineIO (20)
![[DSC Adria 23] Antoni Ivanov Practical Kimball Data Patterns.pptx](https://cdn.slidesharecdn.com/ss_thumbnails/04antoniivanovpracticalkimballdatapatterns-230530184511-e8fd16e8-thumbnail.jpg?width=560&fit=bounds)
Ad
More from Data Con LA (20)
Recently uploaded (20)
Big Data Day LA 2016/ Hadoop/ Spark/ Kafka track - Real-time Aggregations, Approxs, Similarities, and Recommendations at Scale using Spark Kafka, Docker, CoreNLP, Word2Vec, LDA, and Twitter Algebird - Chris Fregly, Research Scientist, PipelineIO
- 1. Extend Your ML Pipelines to Serve Production Pipeline.IOExtend Your ML Pipelines to Serve Production Pipeline.IO advancedspark.compipeline.io
- 2. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Who Am I? 2 Streaming Data Engineer Netflix OSS Committer Data Solutions Engineer Apache Contributor Principal Data Solutions Engineer IBM Technology Center Meetup Organizer Advanced Meetup Book Author Advanced . Due Soon
- 3. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Advanced Apache Spark Meetup http://pipeline.io Meetup Metrics Top 10 Most-active Spark Meetup! ~4000 Members in just 12 mos!! 5100+ Docker downloads (demos) Meetup Mission Code deep-dive into Spark and related open source projects Surface key patterns and idioms Focus on distributed systems, scale, and performance 3
- 4. Extend Your ML Pipelines to Serve Production Pipeline.IOExtend Your ML Pipelines to Serve Production Pipeline.IO Live, Interactive Demo! Audience Participation Required!! Cell Phone Compatible!!! 4
- 5. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production http://demo.pipeline.io End User -> NetflixOSS à Redis à TensorFlow -> Data Scientist -> 5 <- Kafka <- Spark Streaming <- Cassandra Redis <- Zeppelin iPython
- 6. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Presentation Outline ① Scaling ② Similarities ③ Recommendations ④ Approximations ⑤ Netflix Recommendations 6
- 7. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Scaling with Parallelism 7 Peter O(log n) O(log n) Worker Nodes
- 8. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Parallelism with Composability Worker 1 Worker 2 Max (a max b max c max d) == (a max b) max (c max d) Set Union (a U b U c U d) == (a U b) U (c U d) Addition (a + b + c + d) == (a + b) + (c + d) Multiply (a * b * c * d) == (a * b) * (c * d) 8 What about Division and Average? Collect at Driver
- 9. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production What about Division? Division (a / b / c / d) != (a / b) / (c / d) (3 / 4 / 7 / 8) != (3 / 4) / (7 / 8) (((3 / 4) / 7) / 8) != ((3 * 8) / (4 * 7)) 0.134 != 0.857 9 What were the Egyptians thinking?! Not Composable “Divide like an Egyptian”
- 10. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production What about Average? Overall AVG (3, 1) (3 + 5 + 5 + 7) 20 + (5, 1) == -------------------- == --- == 5 + (5, 1) (1 + 1 + 1 + 1) 4 + (7, 1) 10 values counts Pairwise AVG (3 + 5) (5 + 7) 8 12 20 ------- + ------- == --- + --- == --- == 10 != 5 2 2 2 2 2 Divide, Add, Divide? Not Composable Single-Node Divide at the End? Doesn’t need to be Composable! AVG (3, 5, 5, 7) == 5 Add, Add, Add? Composable!
- 11. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Presentation Outline ① Scaling ② Similarities ③ Recommendations ④ Approximations ⑤ Netflix Recommendations 11
- 12. Extend Your ML Pipelines to Serve Production Pipeline.IOExtend Your ML Pipelines to Serve Production Pipeline.IO Similarities 12
- 13. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Euclidean Similarity Exists in Euclidean, flat space Based on Euclidean distance Linear measure Bias towards magnitude 13
- 14. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Cosine Similarity Angular measure Adjusts for Euclidean magnitude bias Normalize to unit vectors in all dimensions Used with real-valued vectors (versus binary) 14 org.jblas. DoubleMatrix
- 15. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Jaccard Similarity Set similarity measurement Set intersection / set union Bias towards popularity Works with binary vectors 15
- 16. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Log Likelihood Similarity Adjusts for popularity bias Netflix “Shawshank” problem 16
- 17. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Word Similarity Edit Distance Misspellings and autocorrect Word2Vec Similar words are defined by similar contexts in vector space 17 English Spanish
- 18. Extend Your ML Pipelines to Serve Production Pipeline.IOExtend Your ML Pipelines to Serve Production Pipeline.IO Demo! Find Synonyms with Word2Vec 18
- 19. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Find Synonyms using Word2Vec 19
- 20. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Document Similarity TF/IDF Term Freq / Inverse Document Freq Used by most search engines Doc2Vec Similar documents are determined by similar contexts 20
- 21. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Bonus! Text Rank Document Summary Text Rank (aka Sentence Rank) Surface summary sentences TF/IDF + Similarity Graph + PageRank Most similar sentence to all other sentences TF/IDF + Similarity Graph Most influential sentences PageRank 21
- 22. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Similarity Pathways (Recommendations) Best recommendations for 2 (or more) people “You like Max Max. I like Message in a Bottle. We might like a movie similar to both.” Item-to-Item Similarity Graph + Dijkstra Heaviest Path 22
- 23. Extend Your ML Pipelines to Serve Production Pipeline.IOExtend Your ML Pipelines to Serve Production Pipeline.IO Demo! Similarity Pathway for Movie Recommendations 23
- 24. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Load Movies with Tags into DataFrame 24 My Choice Their Choice
- 25. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Item-to-Item Tag Jaccard Similarity Based on Tags 25 Calculate Jaccard Similarity (Tag Set Similarity) Must be Above the Given Jaccard Similarity Threshold
- 26. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Item-to-Item Tag Similarity Graph 26 Edge Weights == Jaccard Similarity (Based on Tag Sets)
- 27. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Use Dijkstra to Find Heaviest Pathway 27
- 28. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Calculating Exact Similarity Brute-Force Similarity Cartesian Product O(n^2) shuffle and compute aka. All-pairs, Pair-wise, Similarity Join 28
- 29. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Calculating Approximate Similarity Goal: Reduce Shuffle Approximate Similarity Sampling Bucketing or Clustering Ignore low-similarity probability Locality Sensitive Hashing Twitter Algebird MinHash 29 Bucket By Genre
- 30. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Presentation Outline ① Scaling ② Similarities ③ Recommendations ④ Approximations ① Netflix Recommendations 30
- 31. Extend Your ML Pipelines to Serve Production Pipeline.IOExtend Your ML Pipelines to Serve Production Pipeline.IO Recommendations 31
- 32. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Basic Terminology User: User seeking recommendations Item: Item being recommended Explicit User Feedback: user knows they are rating or liking, can choose to dislike Implicit User Feedback: user not explicitly aware, cannot dislike (click, hover, etc) Instances: Rows of user feedback/input data Overfitting: Training a model too closely to the training data & hyperparameters Hold Out Split: Holding out some of the instances to avoid overfitting Features: Columns of instance rows (of feedback/input data) Cold Start Problem: Not enough data to personalize (new) Hyperparameter: Model-specific config knobs for tuning (tree depth, iterations) Model Evaluation: Compare predictions to actual values of hold out split Feature Engineering: Modify, reduce, combine features Loss Function: Function we’re trying to minimize such as least-squared error for Linear Regression Cross Entropy: Loss function used for classification algorithms such as Logistic Regression Optimizer: Technique to optimize loss function such as Stochastic Gradient Descent (SGD) 32
- 33. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Stochastic Gradient Descent (SGD) Optimizes Loss Function Least Squared Error b/w predicted and actual value Cross Entropy Log Likelihood b/w predicted and actual probability 33 2-Dimensional 3-Dimensional
- 34. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Features Binary: True or False Numeric Discrete: Integers Numeric: Real Values Binning: Convert Continuous into Discrete (Time of Day->Morning, Afternoon) Categorical Ordinal: Size (Small->Medium->Large), Ratings (1->5) Categorical Nominal: Independent, Favorite Sports Teams, Dating Spots Temporal: Time-based, Time of Day, Binge Viewing Text: Movie Titles, Genres, Tags, Reviews (Tokenize, Stop Words, Stemming) Media: Images, Audio, Video Geographic: (Longitude, Latitude), Geohash Latent: Hidden Features within Data (Collaborative Filtering) Derived: Age of Movie, Duration of User Subscription 34
- 35. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Feature Engineering Dimension Reduction Reduce number of features in feature space Principle Component Analysis (PCA) Find principle features that best describe data variance Peel dimensional layers back One-Hot Encoding Convert nominal categorical feature values into 0’s and 1’s Remove any numerical relationship between categories Bears -> 1 Bears -> [1.0, 0.0, 0.0] 49’ers -> 2 --> 49’ers -> [0.0, 1.0, 0.0] Steelers-> 3 Steelers-> [0.0, 0.0, 1.0] 35 Convert Each Item to Binary Vector with Single 1.0 Column
- 36. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Feature Normalization & Standardization Goal Scale features to standard size Prevent boundless features Helps avoid overfitting Required by many ML algos Normalize Features Calculate L1 (or L2, etc) norm, then divide into each element Standardize Features Apply standard normal transformation (mean->0, stddev->1) org.apache.spark.ml.feature.[Normalizer, StandardScaler] 36 http://www.mathsisfun.com/data/standard-normal-distribution.html
- 37. Extend Your ML Pipelines to Serve Production Pipeline.IOExtend Your ML Pipelines to Serve Production Pipeline.IO Non-Personalized Recommendations 37
- 38. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Cold Start Problem “Cold Start” problem New user, don’t know their preferences, must show something! Movies with highest-rated actors Top K aggregations Facebook social graph Friend-based recommendations Most desirable singles PageRank of likes and dislikes 38
- 39. Extend Your ML Pipelines to Serve Production Pipeline.IOExtend Your ML Pipelines to Serve Production Pipeline.IO Demo! GraphFrame PageRank 39
- 40. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Example: Dating Site “Like” Graph 40
- 41. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production PageRank of Top Influencers 41
- 42. Extend Your ML Pipelines to Serve Production Pipeline.IOExtend Your ML Pipelines to Serve Production Pipeline.IO Personalized Recommendations 42
- 43. Extend Your ML Pipelines to Serve Production Pipeline.IOExtend Your ML Pipelines to Serve Production Pipeline.IO Demo! Personalized PageRank 43
- 44. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Personalized PageRank: Outbound Links 44 0.15 = (1 - 0.85 “Damping Factor”) 85% Probability: Choose Among Outbound Network 15% Probability: Choose Self or Random 85% Among Outbound Network
- 45. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Personalized PageRank: No Outbound 45 0.15 = (1 - 0.85 “Damping Factor”) 85% Probability: Choose Among Outbound Network 15% Probability: Choose Self or Random 85% Among No Outbound Network!!
- 46. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production User-to-User Clustering User Similarity Time-based Pattern of viewing (binge or casual) Time of viewing (am or pm) Ratings-based Content ratings or number of views Average rating relative to others (critical or lenient) Search-based Search terms 46
- 47. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Item-to-Item Clustering Item Similarity Profile text (TF/IDF, Word2Vec, NLP) Categories, tags, interests (Jaccard Similarity, LSH) Images, facial structures (Neural Nets, Eigenfaces) Dating Site Example… 47 Cluster Similar Eigen-facesCluster Similar Profiles Cluster Similar Categories
- 48. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Bonus: NLP Conversation Starter Bot 48 “If your responses to my generic opening lines are positive, I may read your profile.” Spark ML, Stanford CoreNLP, TF/IDF, DecisionTrees, Sentiment http://crockpotveggies.com/2015/02/09/automating-tinder-with-eigenfaces.html
- 49. Extend Your ML Pipelines to Serve Production Pipeline.IOExtend Your ML Pipelines to Serve Production Pipeline.IO Bonus: Demo! Spark + Stanford CoreNLP Sentiment Analysis 49
- 50. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Bonus: Top 100 Country Song Sentiment 50
- 51. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Bonus: Surprising Results…?! 51
- 52. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Item-to-Item Based Recommendations Based on Metadata: Genre, Description, Cast, City 52
- 53. Extend Your ML Pipelines to Serve Production Pipeline.IOExtend Your ML Pipelines to Serve Production Pipeline.IO Demo! Item-to-Item-based Recommendations One-Hot Encoding + K-Means Clustering 53
- 54. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production One-Hot Encode Tag Feature Vectors 54
- 55. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Cluster Movie Tag Feature Vectors 55 Hyperparameter Tuning (K Clusters?)
- 56. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Analyze Movie Tag Clusters 56
- 57. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production User-to-Item Collaborative Filtering Matrix Factorization ① Factor the large matrix (left) into 2 smaller matrices (right) ② Lower-rank matrices approximate original when multiplied ③ Fill in the missing values of the large matrix ④ Surface k (rank) latent features from user-item interactions 57
- 58. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Item-to-Item Collaborative Filtering Famous Amazon Paper circa 2003 Problem As users grew, user-to-item collaborative filtering didn’t scale Solution Item-to-item similarity, nearest neighbors Offline (Batch) Generate itemId->List[userId] vectors Online (Real-time) From cart, recommend nearest-neighbors in vector space 58
- 59. Extend Your ML Pipelines to Serve Production Pipeline.IOExtend Your ML Pipelines to Serve Production Pipeline.IO Demo! Collaborative Filtering-based Recommendations 59
- 60. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Fitting the Matrix Factorization Model 60
- 61. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Show ItemFactors Matrix from ALS 61
- 62. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Show UserFactors Matrix from ALS 62
- 63. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Generating Individual Recommendations 63
- 64. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Generating Batch Recommendations 64
- 65. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Clustering + Collaborative Filtering Recs Cluster matrix output from Matrix Factorization Latent features derived from user-item interaction Item-to-Item Similarity Cluster item-factor matrix-> User-to-User Similarity <-Cluster user-factor matrix 65
- 66. Extend Your ML Pipelines to Serve Production Pipeline.IOExtend Your ML Pipelines to Serve Production Pipeline.IO Demo! Clustering + Collaborative Filtering-based Recommendations 66
- 67. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Show ItemFactors Matrix from ALS 67
- 68. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Convert to Item Factors -> mllib.Vector Required by K-Means Clustering Algorithm 68
- 69. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Fit and Evaluate K-Means Cluster Model 69 Measures Closeness Of Points Within Clusters K = 5 Clusters
- 70. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Netflix Genres and Clusters Typical Genres Documentary, Romance, Comedy, Horror, Action, Adventure Latent (Hidden) Clusters Emotionally-Independent Dramas for Hopeless Romantics Witty Dysfunctional-Family TV Animated Comedies Romantic Crime Movies based on Classic Literature Latin American Forbidden-Love Movies Critically-acclaimed Emotional Drug Movie Cerebral Military Movie based on Real Life Sentimental Movies about Horses for Ages 11-12 Gory Canadian Revenge Movies Raunchy Mad Scientist Comedy 70
- 71. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Presentation Outline ① Scaling ② Similarities ③ Recommendations ④ Approximations ⑤ Netflix Recommendations 71
- 72. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production When to Approximate? Memory or time constrained queries Relative vs. exact counts are OK (approx # errors after a release) Using machine learning or graph algos Inherently probabilistic and approximate Streaming aggregations Inherently sloppy collection (exactly once?) 72 Approximate as much as you can get away with! Ask for forgiveness later !!
- 73. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production When NOT to Approximate? If you’ve ever heard the term… “Sarbanes-Oxley” …at the office. 73
- 74. Extend Your ML Pipelines to Serve Production Pipeline.IOExtend Your ML Pipelines to Serve Production Pipeline.IO A Few Good Algorithms 74 You can’t handle the approximate!
- 75. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Common to These Algos & Data Structs Low, fixed size in memory Store large amount of data Known error bounds Tunable tradeoff between size and error Less memory than Java/Scala collections Rely on multiple hash functions or operations Size of hash range defines error 75
- 76. Extend Your ML Pipelines to Serve Production Pipeline.IOExtend Your ML Pipelines to Serve Production Pipeline.IO Bloom Filter Set.contains(key): Boolean “Hash Multiple Times and Flip the Bits Wherever You Land” 76
- 77. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Bloom Filter Approximate Set.contains(key) No means No, Yes means Maybe Elements can only be added Never updated or removed 77
- 78. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Bloom Filter in Action 78 set(key) contains(key): Boolean Images by @avibryant Set.contains(key): TRUE -> maybe contains (other key hashes may overlap) Set.contains(key): FALSE -> definitely does not contain (no key flipped all bits)
- 79. Extend Your ML Pipelines to Serve Production Pipeline.IOExtend Your ML Pipelines to Serve Production Pipeline.IO CountMin Sketch Frequency Count and TopK “Hash Multiple Times and Add 1 Wherever You Land” 79
- 80. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production CountMin Sketch (CMS) Approximate frequency count and TopK for key ie. “Heavy Hitters” on Twitter 80 Matei Zaharia Martin Odersky Donald Trump
- 81. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production CountMin Sketch In Action (TopK Count) 81 Images derived from @avibryant Find minimum of all rows … … Can overestimate, but never underestimate Multiple hash functions (1 hash function per row) Binary hash output (1 element per column) x 2 occurrences of “Top Gun” for slightly additional complexity Top Gun Top Gun Top Gun (x 2) A Few Good Men Taps Top Gun (x 2) add(Top Gun, 2) getCount(Top Gun): Long Use Case: TopK movies using total views add(A Few Good Men, 1) add(Taps, 1) A Few Good Men Taps … … Overlap Top Gun Overlap A Few Good Men
- 82. Extend Your ML Pipelines to Serve Production Pipeline.IOExtend Your ML Pipelines to Serve Production Pipeline.IO HyperLogLog Count Distinct “Hash Multiple Times and Uniformly Distribute Where You Land” 82
- 83. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production HyperLogLog (HLL) Approximate count distinct Slight twist Special hash function creates uniform distribution Hash subsets of data with single, special hash func Error estimate 14 bits for size of range m = 2^14 = 16,384 hash slots error = 1.04/(sqrt(16,384)) = .81% 83 Not many of these
- 84. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production HyperLogLog In Action (Count Distinct) Use Case: Number of distinct users who view a movie 84 0 32 Top Gun: Hour 2 user 2001 user 4009 user 3002 user 7002 user 1005 user 6001 User 8001 User 8002 user 1001 user 2009 user 3005 user 3003 Top Gun: Hour 1 user 3001 user 7009 0 16 UniformDistribution: Estimate distinct # of users by inspecting just the beginning 0 32 Top Gun: Hour 1 + 2 user 2001 user 4009 user 3002 user 7002 user 1005 user 6001 User 8001 User 8002 Combine across different scales user 7009 user 1001 user 2009 user 3005 user 3003 user 3001
- 85. Extend Your ML Pipelines to Serve Production Pipeline.IOExtend Your ML Pipelines to Serve Production Pipeline.IO Locality Sensitive Hashing Set Similarity “Pre-process Items into Buckets, Compare Within Buckets” 85
- 86. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Locality Sensitive Hashing (LSH) Approximate set similarity Pre-process m rows into b buckets b << m; b = buckets, m = rows Hash items multiple times ** Similar items hash to overlapping buckets ** Hash designed to cluster similar items Compare just contents of buckets Much smaller cartesian compare ** Compare in parallel !! Avoids huge cartesian all-pairs compare 86 Chapter 3: LSH
- 87. Extend Your ML Pipelines to Serve Production Pipeline.IOExtend Your ML Pipelines to Serve Production Pipeline.IO DIMSUM Set Similarity “Pre-process and ignore data that is unlikely to be similar.” 87
- 88. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production DIMSUM “Dimension Independent Matrix Square Using MR” Remove vectors with low probability of similarity RowMatrix.columnSimiliarites(threshold) Twitter DIMSUM Case Study 40% efficiency gain over bruce-force Cosine Sim 88
- 89. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Common Tools to Approximate Twitter Algebird Redis Apache Spark 89 Composable Library Distributed Cache Big Data Processing
- 90. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Twitter Algebird Algebraic Fundamentals Parallel Associative Composable Examples Min, Max, Avg BloomFilter (Set.contains(key)) HyperLogLog (Count Distinct) CountMin Sketch (TopK Count) 90
- 91. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Redis Implementation of HyperLogLog (Count Distinct) 12KB per item count 2^64 max # of items 0.81% error Add user views for given movie PFADD TopGun_Hour1_HLL user1001 user2009 user3005 PFADD TopGun_Hour1_HLL user3003 user1001 Get distinct count (cardinality) of set PFCOUNT TopGun_Hour1_HLL Returns: 4 (distinct users viewed this movie) Union 2 HyperLogLog Data Structures PFMERGE TopGun_Hour1_HLL TopGun_Hour2_HLL 91 ignore duplicates Tunable
- 92. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Approximations in Spark Libraries Spark Core countByKeyApprox(timeout: Long, confidence: Double) PartialResult Spark SQL approxCountDistinct(column: Column, targetResidual: Float) approxQuantile(column: Column, quantiles: Seq[Float], targetResidual: Float) Spark ML Stratified sampling sampleByKey(fractions: Map[K, Double]) DIMSUM sampling Probabilistic sampling reduces amount of shuffle RowMatrix.columnSimilarities(threshold: Double) 92
- 93. Extend Your ML Pipelines to Serve Production Pipeline.IOExtend Your ML Pipelines to Serve Production Pipeline.IO Demo! Exact Count vs. Approximate HLL and CMS Count 93
- 94. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production HashSet vs. HyperLogLog (Memory) 94
- 95. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production HashSet vs. CountMin Sketch (Memory) 95
- 96. Extend Your ML Pipelines to Serve Production Pipeline.IOExtend Your ML Pipelines to Serve Production Pipeline.IO Demo! Exact Similarity vs. Approximate LSH Similarity 96
- 97. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Brute Force Cartesian All Pair Similarity 97 47 seconds
- 98. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Locality Sensitive Hash All Pair Similarity 98 6 seconds
- 99. Extend Your ML Pipelines to Serve Production Pipeline.IOExtend Your ML Pipelines to Serve Production Pipeline.IO Many More Demos! or Download Docker Clone on Github 99 http://advancedspark.com
- 100. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Presentation Outline ① Scaling ② Similarities ③ Recommendations ④ Approximations ⑤ Netflix Recommendations 100
- 101. Extend Your ML Pipelines to Serve Production Pipeline.IOExtend Your ML Pipelines to Serve Production Pipeline.IO Netflix Recommendations From Ratings to Real-time 101
- 102. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Netflix Has a Lot of Data Netflix has a lot of data about a lot of users and a lot of movies. Netflix can use this data to buy new movies. Netflix is global. Netflix can use this data to choose original programming. Netflix knows that a lot of people like politics and Kevin Spacey. 102 The UK doesn’t have White Castle. Renamed my favourite movie to: “Harold and Kumar Get the Munchies” My favorite movie: “Harold and Kumar Go to White Castle” Summary: Buy NFLX Stock! This broke my unit tests!
- 103. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Netflix Data Pipeline - Then 103 v1.0 v2.0
- 104. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Netflix Data Pipeline – Now (Keystone) 104 v3.0 9 million events per second 22 GB per second!! EC2 D2XL Disk: 6 TB, 475 MB/s RAM: 30 G Network: 700 Mbps Auto-scaling, Fault tolerance A/B Tests, Trending Now SAMZA Splits high and normal priority
- 105. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Netflix Recommendation Data Pipeline 105 Throw away batch user factors (U) Keep batch video factors (V)
- 106. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Netflix Trending Now (Time-based Recs) Uses Spark Streaming Personalized to user (viewing history, past ratings) Learns and adapts to events (Valentine’s Day) 106 “VHS” Number of Plays Number of Impressions Calculate Take Rate
- 107. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Bonus: Pandora Time-based Recs Work Days Play familiar music User is less likely accept new music Evenings and Weekends Play new music More like to accept new music 107
- 108. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production $1 Million Netflix Prize (2006-2009) Goal Improve movie predictions by 10% (Root Mean Sq Error) Test data withheld to calculate RMSE upon submission 5-star Ratings Dataset (userId, movieId, rating, timestamp) Winning algorithm(s) 10.06% improvement (RMSE) Ensemble of 500+ ML combined with GBDT’s Computationally impractical 108
- 109. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Secrets to the Winning Algorithms Adjust for the following human bias… ① Alice effect: user rates lower than avg ② Inception effect: movie rated higher than avg ③ Overall mean rating of a movie ④ Number of people who have rated a movie ⑤ Number of days since user’s first rating ⑥ Number of days since movie’s first rating ⑦ Mood, time of day, day of week, season, weather 109
- 110. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Netflix Common ML Algorithms Logistic Regression Linear Regression Gradient Boosted Decision Trees Random Forest Matrix Factorization SVD Restricted Boltzmann Machines Deep Neural Nets Markov Models LDA Clustering 110 Ensembles!
- 111. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Netflix Genres and Clusters Typical Genres Documentaries, Romance Comedies, Horror, Action, Adventure Latent (Hidden) Clusters Emotionally-Independent Dramas for Hopeless Romantics Witty Dysfunctional-Family TV Animated Comedies Romantic Crime Movies based on Classic Literature Latin American Forbidden-Love Movies Critically-acclaimed Emotional Drug Movie Cerebral Military Movie based on Real Life Sentimental Movies about Horses for Ages 11-12 Gory Canadian Revenge Movies Raunchy Mad Scientist Comedy 111
- 112. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Netflix Social Integration Post to Facebook after movie start (5 mins) Recommend to new users based on friends Helps with Cold Start problem 112
- 113. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Netflix Search No results? No problem… Show similar results! Utilize extensive DVD Catalog Metadata search (ElasticSearch) Named entity recognition (NLP) Empty searches are opportunity! Explicit feedback for future recommendations Content to buy and produce! 113
- 114. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Netflix A/B Tests Users tend to click on images featuring… Faces with strong emotional expressions Villains over heroes Small number of cast members 114
- 115. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Netflix Recommendation Serving Layer Use Case: Recommendation service depends on EVCache Problem: EVCache cluster goes down or becomes latent!? Answer: github.com/Netflix/Hystrix Circuit Breaker! Circuit States Closed: Service OK Open: Service DOWN Fallback to Static 115
- 116. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Why Higher Average Ratings 2004+? 2004, Netflix noticed higher ratings on average Some possible reasons why… 116 ① Significant UI improvements deployed ② New recommendation engine deployed ③
- 117. Extend Your ML Pipelines to Serve Production Pipeline.IOPipeline.IOExtend Your ML Pipelines to Serve Production Thank You, Everyone!! Chris Fregly @cfregly Research Scientist @ PipelineIO San Francisco, California, USA http://fluxcapacitor.com Sign up for the Meetup and Book Contribute to Github Repo Run all Demos using Docker Find me LinkedIn, Twitter, Github, Email, Fax 117 Image derived from http://www.duchess-france.org/