Guides

Computations

The unit of work that combines algorithms, locations, and time periods

Computations

Computations are the unit of work in the Elements Platform - they represent the actual execution of algorithms or analyses with specific configurations, locations, and time periods.

A computation combines:

  • An Algorithm or Analysis configuration
  • A collection of Areas of Interest (AOIs)
  • A Times of Interest (TOI)

This creates an executable unit that processes specific data to generate results.

Computation

Computation Types

Algorithm Computation

Executes a single algorithm:

Algorithm Config + AOI Collection + TOI = Algorithm Computation

Example: Run vehicle detection on 10 store locations for the past 30 days.

Analysis Computation

Executes a multi-algorithm workflow:

Analysis Config + AOI Collection + TOI = Analysis Computation

Example: Run a 4-algorithm supply chain analysis on shipping routes for Q1 2024.

Computation Lifecycle

1. Creation

  • Define the algorithm/analysis configuration
  • Select AOI collection
  • Specify TOI
  • Submit for execution

2. Queuing

  • Computation enters the processing queue
  • Platform allocates resources
  • Dependencies are resolved (for analyses)

3. Running

  • Data is filtered by AOI and TOI
  • Algorithm containers are launched
  • Processing occurs in parallel (when possible)
  • Progress is tracked and reported

4. Completion

  • Results are validated and stored
  • Status updated to complete
  • Results become available via API or UI

5. Results Access

  • View in Elements Application
  • Download via Elements API
  • Use as input to subsequent analyses

Computation States

Computations progress through various states:

  • Pending: Waiting to start
  • Running: Currently processing
  • Paused: Temporarily stopped (can be resumed)
  • Completed: Successfully finished
  • Failed: Encountered an error
  • Cancelled: Manually stopped by user

Computation Operations

Run

Start a new computation or resume a paused one:

  • Immediate execution or scheduled for later
  • Automatic retry on transient failures
  • Progress tracking available

Pause

Temporarily stop a running computation:

  • Preserves current progress
  • Can be resumed later
  • Useful for resource management

Resume

Continue a paused computation:

  • Picks up where it left off
  • No data reprocessing
  • Maintains result consistency

Cancel

Stop a computation permanently:

  • Cannot be resumed
  • Partial results may be available
  • Resources are released

Delete

Remove a computation and its results:

  • Frees storage space
  • Cannot be undone
  • Use with caution

Parallelization

Computations leverage parallel processing for efficiency:

By Location (AOI)

Each AOI in the collection can be processed independently:

10 AOIs → 10 parallel processing units

By Time (TOI)

Each time period can be processed independently:

30 days → 30 parallel processing units

Combined

Location and time parallelization multiply:

10 AOIs × 30 days = 300 parallel processing units

Result: Dramatically faster completion times for large-scale analyses.

Data Flow in Computations

Algorithm Computation Flow

1. Computation starts
         ↓
2. Platform filters input data by AOI + TOI
         ↓
3. Filtered data divided by AOI and time period
         ↓
4. Algorithm containers launched for each unit
         ↓
5. Containers process data in parallel
         ↓
6. Results written to storage
         ↓
7. Computation marked complete

Analysis Computation Flow

1. Analysis Computation starts
         ↓
2. First algorithm in DAG executes (as above)
         ↓
3. First algorithm results become input to next
         ↓
4. Second algorithm executes using previous results
         ↓
5. Continue through all algorithms in DAG
         ↓
6. Final results stored
         ↓
7. Computation marked complete

Computation History

The platform maintains a history of all computations:

  • Audit trail: Who ran what, when
  • Reproducibility: Re-run with exact same parameters
  • Comparison: Compare results across runs
  • Debugging: Investigate failures or unexpected results

Resource Management

Computations consume platform resources:

CPU and Memory

  • Specified per algorithm
  • Determines container size
  • Affects processing speed

GPU (Optional)

  • For ML/AI algorithms
  • Accelerates certain operations
  • Higher cost but faster processing

Storage

  • Results occupy storage space
  • Larger AOIs/longer TOIs = more storage
  • Cleanup old results to free space

Monitoring and Troubleshooting

Progress Tracking

  • Real-time status updates
  • Percentage complete
  • Estimated time remaining
  • Failed units identified

Logs and Diagnostics

  • Container logs available
  • Error messages captured
  • Debugging information preserved

Performance Metrics

  • Processing time per unit
  • Resource utilization
  • Throughput statistics

Best Practices

Computation Sizing

  • Start small: Test with limited AOIs/TOI first
  • Scale gradually: Increase scope after validation
  • Monitor costs: Larger computations = higher costs

Resource Optimization

  • Use appropriate cadence: Don't over-sample temporally
  • Right-size AOIs: Balance coverage and processing time
  • Clean up: Delete old computations when no longer needed

Error Handling

  • Check logs: Understand failures before retrying
  • Validate inputs: Ensure data exists for AOI/TOI
  • Retry wisely: Transient failures may resolve, but systematic errors won't

Use Cases

One-Time Analysis

  • Run a computation once for a specific question
  • Generate a report or dataset
  • Archive or delete when complete

Recurring Analysis

  • Schedule regular computations (daily, weekly, etc.)
  • Monitor ongoing conditions
  • Track trends over time

Comparative Analysis

  • Run multiple computations with different parameters
  • Compare algorithm versions
  • Test different configurations

Development and Testing

  • Test algorithm changes
  • Validate new analysis workflows
  • Debug processing issues

Next Steps

Updated 5 months ago