O slideshow foi denunciado.
Utilizamos seu perfil e dados de atividades no LinkedIn para personalizar e exibir anúncios mais relevantes. Altere suas preferências de anúncios quando desejar.
Market Propensity Modelling Using
XStreams
Copyright � 2015-2017
Exadatum Software Services Pvt. Ltd.
About XSTREAMS
Why XSTREAMS?
XSTREAMS Architecture
Technology S...
Copyright � 2015-2017
Exadatum Software Services Pvt. Ltd.
XSTREAMS: Drag and Drop Self Serve Platform
50+ Data Processing...
Copyright � 2015-2017
Exadatum Software Services Pvt. Ltd.
Why XSTREAMS?
Sink for Error
Handling
Timeseries and
Aggregated...
Copyright � 2015-2017
Exadatum Software Services Pvt. Ltd.
5
Kafka
Pubsub
Flume
RabbitMQ
Amazon S3
HDFS
Kinesis
MQTT
Data ...
Copyright � 2015-2017
Exadatum Software Services Pvt. Ltd.
6
Source SinkUI/UX
Service Layer
Xstreams Core
Real Time Olap
T...
Copyright � 2015-2017
Exadatum Software Services Pvt. Ltd.
7
.
Marketing Propensity Business Use Case
Predict user purchas...
Copyright � 2015-2017
Exadatum Software Services Pvt. Ltd.
8
.
Overview of Sources/Features/EndPoints
ClickStream Dataset
...
Copyright � 2015-2017
Exadatum Software Services Pvt. Ltd.
9
.
Legacy Modelling Steps
Join ClickStream and Product Dataset...
Copyright � 2015-2017
Exadatum Software Services Pvt. Ltd.
10
.
Legacy Modelling Issues
Pivots creation was taking more th...
Copyright � 2015-2017
Exadatum Software Services Pvt. Ltd.
11
.
Feature Engineering Optimization on Xstreams
Complete Data...
Market Propensity Pipeline on XStreams (Live Demo)
Copyright � 2015-2017
Exadatum Software Services Pvt. Ltd.
Thank You!
13
Próximos SlideShares
Carregando em…5
×

Market Propensity Modeling Using XSTREAMS

Market Propensity Logistic Regression Model Using XSTREAMS: A Unified Batch/Streaming and ML Platform for Big Data.

  • Seja o primeiro a comentar

  • Seja a primeira pessoa a gostar disto

Market Propensity Modeling Using XSTREAMS

  1. 1. Market Propensity Modelling Using XStreams
  2. 2. Copyright � 2015-2017 Exadatum Software Services Pvt. Ltd. About XSTREAMS Why XSTREAMS? XSTREAMS Architecture Technology Stack Adv. Market Propensity Legacy Design & Issues Feature Engineering Modelling on XStreams ??? Agenda
  3. 3. Copyright � 2015-2017 Exadatum Software Services Pvt. Ltd. XSTREAMS: Drag and Drop Self Serve Platform 50+ Data Processing Operators 45+ Transformer and Estimators for Feature Engineering Train, Score and Evaluate Model in one Pipeline Marketplace for readily available blueprints.
  4. 4. Copyright � 2015-2017 Exadatum Software Services Pvt. Ltd. Why XSTREAMS? Sink for Error Handling Timeseries and Aggregated Metrics Actionable Alerts Scheduling Capabilities Auditing Support Versioning Support Granular Role Based Authorization Checkpointing Support. Common features automatically applied to all pipelines created
  5. 5. Copyright � 2015-2017 Exadatum Software Services Pvt. Ltd. 5 Kafka Pubsub Flume RabbitMQ Amazon S3 HDFS Kinesis MQTT Data Sources HDFS Files Hive Kafka ElasticSearch Kinesis WebSocket Cassandra BigQuery PubSub Data Sinks Real Time Self Service ETL Platform Hadoop Distributions Native / VM / Cloud Cloudera MapR HDP High Level Architecture
  6. 6. Copyright � 2015-2017 Exadatum Software Services Pvt. Ltd. 6 Source SinkUI/UX Service Layer Xstreams Core Real Time Olap Technology Stack
  7. 7. Copyright � 2015-2017 Exadatum Software Services Pvt. Ltd. 7 . Marketing Propensity Business Use Case Predict user purchase trends across different signals like Brand, Price , Size based on custom and dynamic feature set that is composed on time based event product category , sub category , age and gender. Business Applications Search and Browse for Recommendation Enhance Browsing Experience Discount Optimization Campaign Management Ad Monetization Internal R &D (Eg. Acti Mirrors) Futuristic Shopping Experience Tagging Using App by Finding Customers
  8. 8. Copyright � 2015-2017 Exadatum Software Services Pvt. Ltd. 8 . Overview of Sources/Features/EndPoints ClickStream Dataset CustomerId ProductId EventId Time Product Dataset Product Category Product Sub Category Brand Occasion Age , Gender , Colour , Size Demographic Dataset Gender Household Size Income #Children , Marital Status Education Source Tables and Attributes Features Product Category Product Sub Category Event Type Age Gender Time EndPoints/Singals Brand Price Size Occasion Colour
  9. 9. Copyright � 2015-2017 Exadatum Software Services Pvt. Ltd. 9 . Legacy Modelling Steps Join ClickStream and Product Dataset Pivot and Vectorize Join ClickStream and Demographic Dataset Pivot and Vectorize Merge Above two Vectors Apply Binary Classification Logistic Regression Model
  10. 10. Copyright � 2015-2017 Exadatum Software Services Pvt. Ltd. 10 . Legacy Modelling Issues Pivots creation was taking more than 20 hours on whole dataset. So sampled 5% dataset was used. Default Vector Operator was taking longer time. Modelling was done on subset of the features combinations. Skewed data for Purchase and Non Purchase label.
  11. 11. Copyright � 2015-2017 Exadatum Software Services Pvt. Ltd. 11 . Feature Engineering Optimization on Xstreams Complete Dataset for 30 million customers was used since vectorization time was reduced from 18hours to 3 hours due to custom sparse vectorize operator. Removed skewed data label by redcing the non purchase by 1:10 ratio. Added unknown values for missing demographic values. All feature combination were not used instead of top 20.P1/P2 combination varied 60-2400
  12. 12. Market Propensity Pipeline on XStreams (Live Demo)
  13. 13. Copyright � 2015-2017 Exadatum Software Services Pvt. Ltd. Thank You! 13

×