Scomp Link

scomp-link — End-to-End ML Toolkit

Overview

scomp-link automates the complete ML workflow: data profiling → preprocessing → feature engineering → model selection → training → validation → explainability → monitoring → deployment.

Use scomp-link instead of raw sklearn when you need:

• Zero-code ML via CLI (26 commands)
• Automated model selection based on data characteristics
• Persistent artifacts (.scomp format: model + preprocessor + config + metrics)
• HTML reports with embedded interactive charts
• Production monitoring (drift + anomaly + fairness)
• One-command hyperparameter tuning (Optuna/Halving)

Use raw sklearn when you need:

• Custom model architectures not in the factory
• Fine-grained control over every preprocessing step
• Research workflows requiring full flexibility

Installation

pip install scomp-link

Decision Tree: Which Command to Use

I have data and want to...
├─ Understand it quickly          → scomp-link describe --data file.csv
├─ Full quality report (HTML)     → scomp-link quality --data file.csv --output report.html
├─ Engineer features              → scomp-link engineer --data file.csv --target y --interactions --log-transform
├─ Train a model
│  ├─ Regression                  → scomp-link run --data file.csv --target y --task regression
│  ├─ Classification              → scomp-link run --data file.csv --target y --task classification
│  ├─ Text classification         → scomp-link text --data file.csv --text-col msg --target label
│  ├─ Clustering                  → scomp-link cluster --data file.csv --n-clusters 5
│  └─ Full pipeline from YAML     → scomp-link pipeline --config pipeline.yaml
├─ Tune hyperparameters           → scomp-link tune --data file.csv --target y --task regression --method optuna
├─ Predict with saved model       → scomp-link predict --artifact model.scomp --data new.csv
├─ Validate on test data          → scomp-link validate --artifact model.scomp --data test.csv --target y
├─ Explain model decisions        → scomp-link explain --artifact model.scomp --data test.csv
├─ Monitor production
│  ├─ Drift only                  → scomp-link drift --reference train.csv --current prod.csv
│  ├─ Full monitoring             → scomp-link monitor --reference train.csv --current prod.csv --artifact model.scomp
│  └─ Anomaly detection           → scomp-link anomaly --data prod.csv --methods iforest,lof,tabnet,transformer
├─ Check fairness/bias            → scomp-link fairness --data preds.csv --target y_true --predicted y_pred --sensitive gender
├─ Forecast time series           → scomp-link forecast --data series.csv --column value --horizon 30
├─ Compare models                 → scomp-link compare --artifacts v1.scomp v2.scomp
├─ Generate HTML report           → scomp-link report --data file.csv --output report.html
├─ Serve as REST API              → scomp-link serve --artifact model.scomp --port 8080
├─ Export to ONNX/pickle          → scomp-link export --artifact model.scomp --format onnx
├─ Scaffold a new project         → scomp-link init my_project
├─ Configure branding defaults    → scomp-link init-config
└─ Use declarative >> DSL         → see Pipeline DSL section below

Recommended Workflow

Standard ML Pipeline

# 1. Profile the data
scomp-link describe --data train.csv --format table

# 2. Check data quality
scomp-link quality --data train.csv --output quality_report.html

# 3. Engineer features
scomp-link engineer --data train.csv --target price --interactions --log-transform --output engineered.csv

# 4. Tune hyperparameters
scomp-link tune --data engineered.csv --target price --task regression --method optuna --n-trials 100 --save-artifact best_model.scomp

# 5. Validate on held-out test data
scomp-link validate --artifact best_model.scomp --data test.csv --target price --report validation.html

# 6. Explain
scomp-link explain --artifact best_model.scomp --data test.csv

# 7. Deploy
scomp-link serve --artifact best_model.scomp --port 8080

YAML Pipeline (all-in-one)

# pipeline.yaml
data: train.csv
target: price
task: regression
engineer:
  interactions: true
  log_transform: true
tune:
  method: optuna
  n_trials: 50
validate:
  test_data: test.csv
  report: validation_report.html
save: models/price_model.scomp

scomp-link pipeline --config pipeline.yaml

Pipeline DSL (>> operator)

For declarative, readable pipeline composition (Python API only):

from scomp_link import CleanStep, SelectStep, ModelStep, TrainStep, LogStep

# ML pipeline — lazy, executes on .run()
results = (
    CleanStep(df)
    >> SelectStep("target")
    >> LogStep("before model")          # optional: logs state without side effects
    >> ModelStep("numerical_prediction")
    >> TrainStep("regression", test_size=0.2)
).run()

# Report pipeline
from scomp_link import SectionStep, TableStep, GraphStep, SaveStep
from scomp_link.utils.report_html import ScompLinkHTMLReport

(
    SectionStep("Results")
    >> TableStep(metrics_df, "Metrics")
    >> GraphStep(fig, "Performance Chart")
    >> SaveStep("report.html")
).run(ScompLinkHTMLReport("My Report"))

When to use DSL vs imperative API:

• Use >> when the pipeline is linear and the steps are known upfront
• Use imperative pipe.import_and_clean_data() / pipe.run_pipeline() when you need branching, loops, or conditional logic

Visualization & Reports

scomp-link has three visualization engines. See visualization-guide.md for full details.

Quick Reference

Creating an HTML Report (Python API)

from scomp_link.utils.report_html import ScompLinkHTMLReport
from scomp_link.utils.plotly_utils import histogram, barchart, linechart, area_chart
from scomp_link.utils.highcharts import streamgraphs, calendar_heatmap, calendar_gantt
from scomp_link.utils.rawgraphs import treemap, sankey_diagram, sunburst

# 1. Initialize
report = ScompLinkHTMLReport(title='My Analysis Report')

# 2. Add sections with content
report.open_section("Data Overview")
report.add_title("Dataset Statistics")
report.add_text("Analysis of 10,000 customer records.")
report.add_dataframe(summary_df, "Summary Statistics")
report.close_section()

# 3. Add charts
report.open_section("Distributions")
fig = histogram(df['age'].values, "Age Distribution")
report.add_graph_to_report(fig, "Age Histogram")
report.close_section()

# 4. Add RAWGraphs SVG
report.open_section("Hierarchical View")
svg = treemap(labels, parents, values, "Revenue by Category")
report.add_rawgraphs_to_report(svg, "Revenue Treemap")
report.close_section()

# 5. Add Highcharts
report.open_section("Time Trends")
html_stream = streamgraphs("Sales Trend", dates, series_dict, area=True)
report.html_report += html_stream  # Direct HTML append for Highcharts
report.close_section()

# 6. Save
report.save_html('analysis_report.html')
report.save_pdf('analysis_report.pdf')  # requires Playwright

Available Charts (31 RAWGraphs + 5 Plotly + 3 Highcharts)

Comparisons: barchart, barchartmultiset, barchartstacked, piechart, radarchart, voronoidiagram Distributions: beeswarm, boxplot, violinplot Time Series: bumpchart, gantt_chart, horizongraph, linechart, slopechart, streamgraph Correlations: bubblechart, contour_plot, convex_hull, hexagonal_binning, matrixplot, parallelcoordinates Hierarchies: circlepacking, circular_dendrogram, dendrogram, sunburst, treemap, voronoi_treemap Networks: alluvial_diagram, arc_diagram, chord_diagram, sankey_diagram Plotly: histogram, multiple_histograms, barchart, linechart, area_chart Highcharts: streamgraphs, calendar_heatmap, calendar_gantt

Concatenation Patterns

Commands are designed to chain — the output of one is the input of the next:

describe → (understand columns) → engineer → (engineered.csv) → tune → (model.scomp) → validate → (metrics)
                                                                                      ↓
                                                                               predict (new data)
                                                                                      ↓
                                                                               serve (REST API)

Key patterns:

• --save-artifact model.scomp → use with --artifact model.scomp in predict/validate/explain/export/serve
• --output file.csv → use as --data file.csv in next command
• --report file.html → standalone HTML output (viewable in browser)
• --plot file.html → chart output for forecast/drift/anomaly/cluster/compare

Common Errors and Fixes

MCP Server

scomp-link includes an MCP server (33 tools) for agent integration:

# Start the MCP server (stdio mode for Claude Desktop / Kiro / Cursor)
scomp-link mcp

# Or run directly
python -m scomp_link.mcp_server

Report Builder Workflow (MCP)

For building custom branded HTML reports step-by-step:

1. report_create(title, ...) → returns report_id (uses ~/.scomp-link/config.yaml defaults)
2. report_add_section(report_id, title) → opens collapsible section
3. report_add_text(report_id, content, style) → paragraph/title/subtitle/html
4. report_add_table(report_id, json_data, title) → interactive table
5. report_add_chart(report_id, engine, chart_type, data, title) → 41 chart types (6 plotly + 31 rawgraphs + 3 highcharts + 1 custom)
6. report_add_kpi_cards(report_id, metrics_json, cols) → KPI cards with trend/status
7. report_add_tabs(report_id, tabs_json, title) → tabbed navigation (html/chart/table)
8. report_add_comparison_table(report_id, data, baseline_col, compare_cols, ...) → delta comparison
9. report_add_summary_stats(report_id, data_json, title) → auto data profiling table
10. report_add_dark_mode_toggle(report_id) → floating dark/light toggle
11. report_add_code(report_id, code, language, title, output, line_numbers, collapsed) → syntax-highlighted code block with copy button
12. report_add_diff(report_id, old_code, new_code, language, title, old_label, new_label, collapsed) → side-by-side diff view
13. report_add_mermaid(report_id, diagram, title, collapsed) → Mermaid.js diagram (flowchart, sequence, gantt, etc.)
14. report_add_terminal(report_id, cast_data, title, cols, rows, theme, collapsed) → embedded terminal replay (asciinema)
15. report_save(report_id, output) → saves HTML, frees memory

Engines: plotly (interactive), rawgraphs (SVG static), highcharts (time series) New plotly charts: index_chart, stacked_area_comparison (in addition to histogram, barchart, linechart, area_chart) Config: scomp-link init-config creates ~/.scomp-link/config.yaml with branding defaults

See workflow-patterns.md for complete workflow examples.