Overview

scomp-link is an end-to-end machine learning toolkit that automates the complete ML workflow — from data profiling and preprocessing to model selection, training, validation, explainability, monitoring, and deployment.

It includes a full-featured CLI for zero-code ML workflows and a Python API for programmatic use.

Installation

pip install scomp-link

Requires Python 3.10+. Import is near-instant (~6ms) thanks to lazy loading — heavy dependencies load only when needed. Python 3.14 is supported experimentally (TensorFlow not yet available on 3.14).

Key Features

CLI Quick Start

# Scaffold a new project
scomp-link init my_project

# Quick dataset profiling
scomp-link describe --data data.csv --format table

# Full data quality report
scomp-link quality --data data.csv --output report.html

# Feature engineering
scomp-link engineer --data data.csv --target y --interactions --log-transform --output features.csv

# Train a model
scomp-link run --data features.csv --target y --task regression --save-artifact model.scomp

# Train with text data
scomp-link text --data tickets.csv --text-col message --target category --method contrastive --head auto

# Extract embeddings from trained model
scomp-link embed --data new_texts.csv --text-col message --artifact model.scomp --output embeddings.npy

# Clustering
scomp-link cluster --data customers.csv --n-clusters 5 --plot clusters.html

# Hyperparameter tuning
scomp-link tune --data train.csv --target y --task regression --method optuna --n-trials 100

# Predict
scomp-link predict --artifact model.scomp --data new_data.csv --output predictions.csv

# Validate on test data
scomp-link validate --artifact model.scomp --data test.csv --target y --report report.html

# Explain
scomp-link explain --artifact model.scomp --data test.csv

# Detect drift
scomp-link drift --reference train.csv --current production.csv --plot drift.html

# Production monitoring (drift + quality + performance)
scomp-link monitor --reference train.csv --current prod.csv --artifact model.scomp --target y

# Forecast time series
scomp-link forecast --data series.csv --column value --horizon 30 --plot forecast.html

# Anomaly detection
scomp-link anomaly --data data.csv --methods iforest,lof,tabnet,transformer

# Fairness check
scomp-link fairness --data preds.csv --target y_true --predicted y_pred --sensitive gender

# Compare models
scomp-link compare --artifacts v1.scomp v2.scomp --plot comparison.html

# Run full pipeline from YAML config
scomp-link pipeline --config pipeline.yaml

# Serve model as REST API
scomp-link serve --artifact model.scomp --port 8080

# Export model to standard format
scomp-link export --artifact model.scomp --format onnx

# Generate reports
scomp-link report --data data.csv --output eda_report.html
scomp-link report --artifact model.scomp --data test.csv --output model_report.html

# Utilities
scomp-link list-models
scomp-link check-deps

# Configuration
scomp-link init-config              # Create global config (~/.scomp-link/config.yaml)
scomp-link init-config --local      # Create project-level config (.scomp-link.yaml)

Python API Quick Start

from scomp_link import ScompLinkPipeline, ScompArtifact, set_verbosity
import pandas as pd

# Control output
set_verbosity("info")  # "silent" | "warning" | "info" | "debug"

# Build pipeline
pipe = ScompLinkPipeline("My Project")
pipe.set_objectives(["Minimize RMSE"])
pipe.import_and_clean_data(df)
pipe.select_variables(target_col='target')
pipe.choose_model("numerical_prediction")
results = pipe.run_pipeline(task_type="regression")

# Save as artifact
artifact = ScompArtifact()
artifact.set_model(pipe.model)
artifact.set_config(task_type='regression', target_col='target')
artifact.set_metrics(results['metrics'])
artifact.save('model.scomp')

# Load and predict
loaded = ScompArtifact.load('model.scomp')
predictions = loaded.predict(new_data)

Feature Engineering

from scomp_link import FeatureEngineer

fe = FeatureEngineer(
    interactions=True,      # Polynomial interactions
    log_transform=True,     # Log1p for skewed features
    date_features=True,     # Extract year/month/dow/weekend
    target_encode=True,     # Encode high-cardinality categoricals
    auto_bin=True,          # Quantile binning
)
X_train_eng = fe.fit_transform(X_train, y_train)
X_test_eng = fe.transform(X_test)

Advanced Hyperparameter Tuning

from scomp_link.models.advanced_tuning import OptunaOptimizer

def param_space(trial):
    return {
        'n_estimators': trial.suggest_int('n_estimators', 50, 500),
        'max_depth': trial.suggest_int('max_depth', 3, 20),
        'learning_rate': trial.suggest_float('learning_rate', 0.01, 0.3, log=True),
    }

optimizer = OptunaOptimizer(GradientBoostingRegressor, param_space, scoring='r2', n_trials=100)
best_model = optimizer.optimize(X_train, y_train)

Explainability

from scomp_link import ShapExplainer, LimeExplainer

# SHAP
shap_exp = ShapExplainer(model, X_train[:100])
shap_exp.explain(X_test)
importance = shap_exp.feature_importance()
fig = shap_exp.plot_importance()

# LIME
lime_exp = LimeExplainer(model, X_train, task='regression')
exp = lime_exp.explain_instance(X_test.iloc[0])
fig = lime_exp.plot_explanation(exp)

Data Drift Detection

from scomp_link import DriftDetector

detector = DriftDetector(X_train, psi_threshold=0.2)
report = detector.detect(X_production)
summary = detector.summary(report)
fig = detector.plot_drift_report(report)

Fairness & Bias Metrics

from scomp_link import FairnessMetrics

fm = FairnessMetrics(y_true, y_pred, sensitive_feature=df['gender'])
report = fm.compute_all()
print(fm.summary(report))
fig = fm.plot_fairness_report(report)

Time Series Forecasting

from scomp_link import TimeSeriesForecaster

fc = TimeSeriesForecaster(method='auto', horizon=30)
fc.fit(series)
forecast = fc.predict_with_ci()
cv_results = fc.walk_forward_cv(series, n_splits=5)
fig = fc.plot_forecast()

Data Quality Report

from scomp_link import DataQualityReport

dqr = DataQualityReport(df)
report = dqr.generate()  # missing, cardinality, constants, duplicates, correlations
dqr.save_html('quality_report.html')

Anomaly Detection

from scomp_link import AnomalyDetector

detector = AnomalyDetector(
    contamination=0.05,
    methods=['iforest', 'lof', 'tabnet', 'transformer'],
    consensus_threshold=2,
)
results = detector.fit_predict(df, features=['col1', 'col2', 'col3'])

Project Structure

scomp_link/
├── cli.py                    # CLI (24 commands)
├── core.py                   # ScompLinkPipeline orchestrator
├── preprocessing/
│   ├── data_processor.py     # Preprocessor (polars backend)
│   ├── feature_engineer.py   # FeatureEngineer (sklearn-compatible)
│   └── data_quality.py       # DataQualityReport
├── models/
│   ├── model_factory.py      # Decision-tree model selection
│   ├── regressor_optimizer.py
│   ├── classifier_optimizer.py
│   ├── ensemble_optimizer.py
│   ├── advanced_tuning.py    # Optuna, Halving, EarlyStopping
│   ├── forecaster.py         # TimeSeriesForecaster
│   ├── anomaly_detector.py
│   ├── ts_anomaly_detector.py
│   ├── contrastive_text.py   # BERT contrastive learning
│   ├── supervised_text.py
│   └── supervised_img.py
├── validation/
│   ├── model_validator.py    # Metrics + HTML reports
│   ├── advanced_cv.py        # LOOCV, Bootstrap
│   └── fairness.py           # FairnessMetrics
├── explainability/
│   └── explainer.py          # ShapExplainer, LimeExplainer
├── monitoring/
│   └── drift_detector.py     # DriftDetector (PSI + KS)
├── persistence/
│   └── artifact.py           # ScompArtifact (.scomp format)
└── utils/
    ├── colors.py             # Centralized color palettes
    ├── logger.py             # Configurable logging
    ├── report_html.py        # HTML report builder
    ├── plotly_utils.py       # Plotly chart utilities
    ├── highcharts.py         # Highcharts visualizations
    └── rawgraphs/            # 31 SVG chart functions (server-side)

AI Agent Integration

scomp-link works natively with AI agents via **MCP (Model Conte