Skip to main content

FLNet Data Analysis

FLNet Data Analysis is an enterprise-grade environment for running reproducible, tool-based analyses on structured and semi-structured data. It combines:

  • Data management (upload, preview, organize inputs and outputs)
  • A workflow canvas to build DAG pipelines
  • A Tool Store for algorithms, transformations, and trainable models
  • Asynchronous execution with persisted results
  • LLM-assisted result analysis (on-demand)
  • PDF reporting for sharing and reproducibility
  • Strict user isolation (each user sees only their own assets)

This documentation explains the platform end-to-end so you can use it without external guidance.

Mental model: how it works​

In FLNet, everything follows the same lifecycle:

  1. Upload or select data (inputs)
  2. Build a workflow (DAG of tool nodes)
  3. Configure inputs, features, and hyperparameters
  4. Run (asynchronous job execution)
  5. Inspect outputs (files, plots, reports, artifacts)
  6. Optional: Analyze results with LLM
  7. Export a PDF report for sharing and reproducibility

Core concepts​

User isolation (enterprise baseline)​

FLNet enforces strict separation between users:

  • You can access only your own files, workflows, runs, and reports.
  • Sharing is done through exported artifacts (e.g., PDF report), not by exposing your workspace to others.

Workflows (DAG pipelines)​

A workflow is a graph-based pipeline (a DAG) composed of tool nodes.

  • Nodes represent tools/models/transformations.
  • Edges represent execution order and dependencies.
  • Workflows are saved and reusable.
  • The platform currently executes the latest workflow version.

Execution mode depends on deployment:

  • Local: executed in a local environment
  • Global: executed in a shared/global environment (e.g., centralized compute)

Note: The UI presents workflows as a graph. Internally, FLNet resolves the DAG into a valid execution order.

Tools and tool types​

Tools come from the Tool Store and can include:

  • Trainable models (produce model artifacts)
  • Clustering tools (produce reports and visualizations)
  • File transformation tools (reshape, filter, convert data)
  • Custom tools specific to your deployment

Each tool defines:

  • required/optional inputs
  • typed outputs
  • typed hyperparameters (configuration)
  • accepted file types (may be fixed or mixed, depending on tool implementation)

UI layout and navigation​

FLNet Data Analysis is typically split into three working areas:

  1. Left panel: Data Management

    • Upload files
    • Browse inputs
    • Browse outputs created by runs

  2. Center: Workflow area

    • Build and configure workflows (DAG)
    • Open Tool Store and add tools/models
    • Start executions and access results

  3. Top-right toolbar

    • Export a PDF report documenting the full analysis (inputs, parameters, outputs)

Data Management​

Uploading files​

In the Data Management section you can upload new files for analysis.

What you can upload depends on your deployment, but typically includes:

  • CSV and other tabular formats
  • PDFs
  • Images (PNG/JPG)
  • Tool-specific formats (custom)

Mixed/advanced types are supported when a tool explicitly implements them.

Opening file details​

Click any file to open its file detail view. This is the main place to:

  • validate that inputs look correct
  • understand structure and columns
  • preview outputs created by tools

File previews by type​

FLNet shows specialized previews depending on file type:

  • CSV: Table preview + configurable diagrams
  • PDF: PDF viewer
  • Images: image viewer
  • Other formats: preview depends on what the tool provides (may be download-only)

CSV detail view (Table + Diagram)​

When opening a CSV (e.g., iris.csv), you typically see:

Table tab​

  • Column headers and rows in a spreadsheet-like preview
  • Fast verification of schema and values (e.g., numeric features, label columns)

Common usage:

  • Confirm required columns exist
  • Check missing values / obvious data issues
  • Identify which columns are features vs. labels/metadata

Diagram tab​

The Diagram view is configurable and designed for lightweight EDA. Typical actions (depending on enabled diagram types):

  • Select columns for x/y axes
  • Choose aggregation/grouping
  • Compare distributions or relationships

Use Diagram to validate assumptions before running a workflow (e.g., scale, outliers, separability).

Outputs: where results appear​

Every tool run can produce output artifacts. These show up in:

  • the Output section in the left panel
  • the result view for the executed tool/workflow (depending on UI configuration)

Outputs can include:

  • plots (e.g., dendrogram.png)
  • text reports
  • transformed datasets
  • trained model artifacts
  • additional intermediate artifacts (tool-dependent)

Opening outputs works exactly like opening inputs: click the file to open its detail view.

Tool Store: add capabilities to workflows​

What the Store contains​

The Store is your catalog of runnable building blocks:

  • tools (algorithms, transformations)
  • trainable models
  • prebuilt workflows (if provided by deployment)

Adding a tool/model to the canvas​

From the workflow area:

  1. Open the Store
  2. Select a tool/model
  3. Add it to the workflow canvas
  4. Connect nodes if required (DAG dependencies)

Configure a tool node (inputs, features, hyperparameters)​

When selecting a tool node (example: AgglomerativeClustering), you configure:

1) Inputs​

  • Choose one or more input files (e.g., a CSV dataset)

2) Features​

  • Select which columns should be used as model features
  • Feature selection is typically provided as a dropdown or multi-select

Best practice:

  • Exclude label columns (e.g., VARIETY) unless the tool explicitly expects them.
  • Prefer numeric columns for most ML algorithms unless categorical support is explicitly stated.

3) Hyperparameters (typed)​

FLNet tools expose hyperparameters in a typed form.

  • Values are validated against the tool specification.
  • Tools may show a compact summary (e.g., “10 Hyperparams”) with a dedicated editor.

Examples of typed inputs:

  • number fields (int/float)
  • enum selections
  • booleans
  • structured configs (tool-dependent)

If a tool supports mixed/custom file types, it is responsible for correctly interpreting them.

Running analyses (asynchronous execution)​

FLNet runs tools and workflows asynchronously:

  • You start an execution.
  • The platform schedules and runs it in the configured environment.
  • Outputs appear once the run completes.

This is essential for enterprise scenarios:

  • long-running jobs do not block the UI
  • results are persisted as artifacts
  • reproducibility is supported through stored configurations and reports

If a run fails:

  • the tool/run status reflects the failure (deployment dependent)
  • the output may contain logs or error artifacts (tool dependent)

Result review: reports and visualizations​

After completion, review:

  • Text reports (high-level outcome summaries)
  • Plots and figures (e.g., dendrograms, metrics charts)
  • Generated datasets (transformed data, predictions)
  • Model artifacts (trained models, weights, configs)

Example: Agglomerative Clustering

  • A run may produce:
    • a short report (e.g., number of clusters)
    • a dendrogram image

LLM-assisted result analysis (on-demand)​

FLNet can analyze result artifacts with an LLM.

How it works​

LLM analysis is manual:

  • You open a result file (e.g., dendrogram.png)
  • You click Analyze
  • FLNet generates a textual analysis

What you get (standard structure)​

The analysis is returned as text and typically includes:

  1. What the file shows

    • objective description of the artifact (plot type, axes, structure)

  2. Interpretation in context

    • ties results back to the selected tool, features, and goal

  3. Quality / plausibility assessment

    • sanity checks and consistency notes
    • highlights if the result is plausible for the given setup

  4. Notable patterns / anomalies

    • potential artifacts, limitations, suspicious patterns

When to use it​

  • To speed up interpretation of visual outputs
  • To produce a first-pass scientific explanation
  • To document results consistently across teams

The LLM is an assistant, not a replacement for domain validation. Use it for guidance and structured review.

Chat interface (multi-agent assistant)​

FLNet includes a chat assistant designed for end-to-end workflow support.

Capabilities​

The assistant can do everything the user can do in the UI, including:

  • recommend suitable tools/models
  • explain features and parameters
  • guide workflow construction (DAG)
  • help troubleshoot configurations
  • provide result interpretation guidance

Context access (within your workspace)​

The assistant can reference:

  • your uploaded files
  • your workflows
  • your runs and output artifacts
  • your generated reports

Because of strict user isolation, it only operates on your own data and assets.

PDF reporting (shareable reproducibility artifact)​

The PDF button in the top-right toolbar creates a shareable report that documents your analysis.

Purpose​

  • Share with colleagues
  • Provide a reproducible record of what was run
  • Support internal review, validation, and handover

What the report includes​

The PDF report typically captures:

  • tools used (and their order if workflow-based)
  • hyperparameter settings (typed values)
  • selected input files
  • generated output files
  • run metadata (timestamps and other identifiers, deployment dependent)

Depending on configuration, it may also include:

  • thumbnails/previews of figures
  • summary text outputs
  • references to result artifacts

Treat the PDF report as the canonical “analysis snapshot” for collaboration.

  1. Upload inputs

    • Verify in file detail view (CSV Table, PDF/image preview)

  2. Explore (optional)

    • Use CSV Diagram tab for quick EDA

  3. Build workflow

    • Add tools/models from the Store
    • Connect nodes into a DAG

  4. Configure

    • Select inputs
    • Select features
    • Set hyperparameters

  5. Run

    • Start execution (async)

  6. Review outputs

    • Open result artifacts (reports, plots, datasets)

  7. Analyze results (optional)

    • Click Analyze on key artifacts

  8. Export PDF

    • Generate a report for sharing and reproducibility

FAQ​

Which users can see my files and results?​

Only you. FLNet enforces strict isolation: users see only their own assets.

Does the platform run workflows synchronously?​

No. Executions are asynchronous.

Are tool parameters validated?​

Yes. Hyperparameters are typed and validated per tool definition.

Are LLM analyses generated automatically?​

No. LLM analysis is created only when you click Analyze on a result artifact.

Can apps handle custom/mixed file types?​

Yes, if the tool supports them. Tools define how they interpret and validate such inputs, Files can be set as Path

Can I run locally and globally?​

Yes—depending on deployment configuration. The same workflow concept applies in both modes.

Glossary​

  • DAG: Directed Acyclic Graph; a pipeline graph without cycles.
  • Tool Store: Catalog of tools/models/workflows available to add to the canvas.
  • Tool Node: A runnable unit within a workflow (algorithm/model/transformation).
  • Hyperparameters: Typed configuration values controlling tool execution behavior.
  • Run: One asynchronous execution of a tool/workflow that produces outputs.
  • Artifact: Any produced output file (plots, datasets, reports, model files).
  • LLM Analysis: On-demand textual interpretation of a selected result artifact.
  • PDF Report: Shareable reproducibility document capturing tools, parameters, inputs, and outputs.