Skip to contents

NLDR Visualization Tool Server Logic

Source: R/server.R
nldr_viz_server.Rd

Implements the comprehensive server-side logic for the NLDR visualization tool. This function handles all reactive computations, data processing, visualization generation, and user interactions for the multi-tab NLDR analysis application.

Usage

nldr_viz_server(input, output, session)

Arguments

input

The shiny input object containing all user interface inputs. This includes form controls, button clicks, plot selections, and file uploads.

output

The shiny output object for sending rendered content to the UI. Used for plots, tables, text outputs, and dynamic UI elements.

session

The shiny session object for managing client-server communication. Provides access to session state, input updates, and client information.

Value

Invisible NULL. The function sets up reactive expressions and observers that handle all server-side logic. The actual outputs are managed through the shiny reactive system and sent to the client via the output object.

Details

The server function manages several key areas of functionality:

Data Management:

  • File upload validation and CSV parsing with error handling

  • Built-in dataset loading (four_clusters, pdfsense)

  • Empty cell detection and data quality validation

  • Column selection and filtering capabilities

  • Dynamic dataset storage and retrieval system

NLDR Computations:

  • t-SNE implementation with parameter validation and auto-adjustment

  • UMAP processing with neighbor and distance parameter controls

  • Asynchronous computation using future package for responsiveness

  • Progress tracking and user feedback during long computations

  • Result caching and session management

Interactive Visualizations:

  • Plotly-based interactive scatter plots with zoom, pan, and selection

  • Color mapping with automatic palette generation

  • Linked brushing across multiple visualizations

  • Responsive plot sizing and layout management

  • Hover tooltips and selection feedback

Dynamic Tours:

  • Integration with detourr package for animated projections

  • Multiple display types: Scatter, Sage, and Slice projections

  • 5-nearest neighbor graph construction and visualization

  • Real-time parameter adjustment and tour customization

  • Coordinated views between static and dynamic visualizations

Quality Assessment (Quollr Integration):

  • Automated binwidth optimization using RMSE minimization

  • Hexagonal binning and centroid extraction

  • High-dimensional model fitting and validation

  • Prediction error analysis and visualization

  • 3D model tours using langevitour integration

Method Comparison:

  • Side-by-side visualization comparison with linked brushing

  • RMSE-based parameter optimization comparison

  • Best configuration identification and reporting

  • Interactive comparison plot generation

State Management:

  • Reactive value system for maintaining application state

  • Session-based data persistence and cleanup

  • Asynchronous operation tracking and user feedback

  • Memory management for large datasets

Parallel Processing: The server automatically configures parallel processing:

  • Uses multicore on supported systems, multisession otherwise

  • Configures 2 worker processes for optimal performance

  • Proper cleanup on session end to prevent memory leaks

  • Future-based asynchronous computation for UI responsiveness

Note

This function requires several packages to be available:

  • Core: shiny, magrittr

  • Visualization: plotly, ggplot2, DT, scales

  • NLDR: Rtsne, umap, FNN

  • Quality: quollr (with all its dependencies)

  • Tours: detourr, tourr

  • Data: dplyr, crosstalk

  • Async: future

  • Utils: stats, utils, tools

Reactive Values

The server maintains several key reactive values:

  • dataset: Current active dataset

  • vis_results: NLDR computation results

  • shared_vis_data: Crosstalk-enabled data for linked brushing

  • nldr_datasets: Storage for multiple NLDR results

  • optimal_config: Best binwidth configuration from optimization

  • quollr_results: Quality assessment results

  • color_palette: Current color scheme for visualizations

  • is_running_*: Boolean flags for operation status tracking

Error Handling

Comprehensive error handling includes:

  • File upload validation with user-friendly error messages

  • NLDR computation error catching with fallback options

  • Memory management for large datasets

  • Network timeout handling for async operations

  • Graceful degradation when optional features are unavailable

Performance Considerations

  • Asynchronous computations prevent UI blocking

  • Efficient data structures for large datasets

  • Caching of expensive computations

  • Memory cleanup and garbage collection

  • Optimized reactive dependency management

See also

Author

GSoC Contributor