Provenance for Interactive Visualizations Fotis Psallidas Eugene Wu fotis@cs.columbia.edu ewu@cs.columbia.edu
Provenance Primer Fine-Grained Provenance (Connections between input and output tuples)
Provenance Primer Fine-Grained Provenance (Connections between input and output tuples) O = γ #$%$&,%()(+&,%-) (Airports ⨝ Flights) Airports name = from name state a 1 LGA NY O a 2 JFK NY from delay state a 3 IAH TX state avg(delay) j 1 LGA 30 NY γ ⨝ j 2 LGA 40 NY O 1 NY 40 j 3 JFK 50 NY O 2 TX 60 Flights j 4 IAH 60 TX from delay f 1 LGA 30 f 2 LGA 40 f 3 JFK 60 f 4 IAH 60
Provenance Primer Fine-Grained Provenance (Connections between input and output tuples) Airports name state a 1 LGA NY O a 2 JFK NY from delay state a 3 IAH TX state avg(delay) j 1 LGA 30 NY j 2 LGA 40 NY O 1 NY 40 j 3 JFK 50 NY O 2 TX 60 Flights j 4 IAH 60 TX from delay f 1 LGA 30 f 2 LGA 40 {f 4 } = backward_trace({o 2 } , Flights ) f 3 JFK 60 f 4 IAH 60
Provenance Primer Fine-Grained Provenance (Connections between input and output tuples) Airports name state a 1 LGA NY O a 2 JFK NY from delay state a 3 IAH TX state avg(delay) j 1 LGA 30 NY j 2 LGA 40 NY O 1 NY 40 j 3 JFK 50 NY O 2 TX 60 Flights j 4 IAH 60 TX from delay f 1 LGA 30 f 2 LGA 40 {o 2 } = forward_trace({f 4 } , O ) f 3 JFK 60 f 4 IAH 60
Provenance Primer Fine-Grained Provenance (Connections between input and output tuples) Airports name state a 1 LGA NY O a 2 JFK NY from delay state a 3 IAH TX state avg(delay) j 1 LGA 30 NY j 2 LGA 40 NY O 1 NY 40 j 3 JFK 50 NY O 2 TX 60 Flights j 4 IAH 60 TX from delay f 1 LGA 30 f 2 LGA 40 f 3 JFK 60 f 4 IAH 60
Provenance Primer Fine-Grained Provenance (Connections between input and output tuples) • Navigation of the input-output connections • {records} = backward_trace(…) • {records} = forward_trace(…) • Provenance consuming queries • SQL(backward_trace(…)) • SQL(forward_trace(…))
Goal of this talk How to use fine-grained provenance to express core interactive application functionality Why though? q (Expressivity) Logic over provenance is expressed declaratively q (Performance) Provenance management systems are becoming *fast* See [Smoke, VLDB18] or pass by our demo on Wednesday/Thursday
Connections Core interaction logic with provenance • Selections • Logic over selections • Multi-view linking
Interactive Selections Goal: Get subset of inputs that correspond to selected visual outputs Example: Find the airports that operate at the selected states airports View V 1 flights
Interactive Selections Goal: Get subset of inputs that correspond to selected visual outputs Example: Find the airports that operate at the selected states airports = backward_trace( ,airports) airports View V 1 flights
Logic over Selections Goal: Express application logic over the selected inputs Example: Find the # airports that operate at the selected states SQL(backward_trace( ,airports)) 54 airports operate in this area V 2 airports View V 1 flights
Multi-View linking Goal: Look at the relationships between different views Example: Show the distribution of #flights per carrier only for selected states backward_trace( ,flights) airports View V 1 flights airlines View V 2 selective_refresh(V2, flights ) Carrier
Provenance for Interactive Visualizations • Generalized selections } Interactive Selections • Item selection backward_trace(…) • Group selection • Range selection • Semantic Zooming } Logic over Selections • Tooltips SQL(backward_trace(…)) • Details-On-Demand } Multi-View Linking • Linked Brushing selective_refresh(backward_trace(…)) • Crossfilter
What next? Traditionally provenance systems have been at the core of several applications Network Resource … Auditing Data Integration Debugging Diagnostics Scheduling Interactive Interactive Interactive Multi-Application Multi-Application What-if Query Visualizations Data Profiling Linking Linking Provisioning Specification Why-not Viz Workflow Interactive Query Iterative Analytics Debugging Data Cleaning … Explanations Analytics Visualization ML Interaction Interaction Application Deconstruction Interpretability Debugging By Example Design Search and Restyling Replication and Collaborative Action Recovery Sense-Making Meta-Analysis Reproducibility Communication
What next? Traditionally provenance systems have been at the core of several applications Resource Network … Data Integration Debugging Auditing Diagnostics Scheduling (Fast) Provenance management systems can make a difference on several other domains Interactive Interactive Multi-Application Interactive What-if Query Visualizations Data Profiling Linking Provisioning Specification Query Why-not Iterative Viz Workflow Interactive Analytics Debugging Data Cleaning Explanations Analytics … Visualization Interaction ML Interaction Application Deconstruction Debugging Interpretability By Example Design Search and Restyling Collaborative Replication and Action Recovery Sense-Making Meta-Analysis Communication Reproducibility
Multi-Application Linking • Many applications are built over the same database (esp. in enterprises) • Extend multi-view linking to multi-application linking • Powerful for: connecting data across apps, reuse app logic Vis App Data Store Search App HOUSTON airports 2:50-5:60 DL33 V 1 V 2 airlines IAH, TX 3:50-6:60 UA22 flights IAH, TX
Multi-Application Linking • Many applications are built over the same database (esp. in enterprises) • Extend multi-view linking to multi-application linking • Powerful for: connecting data across apps, reuse app logic Vis App Data Store Search App Average Departure Delay 60 minutes HOUSTON airports 2:50-5:60 DL33 V 1 V 2 airlines IAH, TX 3:50-6:60 UA22 flights IAH, TX
Takeaway Interaction Logic as Provenance ⇓ Declarative wins & Holistic optimization
Thank You //Q
Recommend
More recommend
