scomp-link Demo Report

CLI Command

scomp-link describe --data train.csv --format table

▶ Output

  column  dtype  missing%  unique   min     max        mean        std
    sqft  int64     0.0%     224   452    3041   1498.5417   388.3374
bedrooms  int64     0.0%       4     2       5      3.4500     1.1917
     age  int64     0.0%      49     1      49     24.5500    14.4864
  garage  int64     0.0%       3     0       2      1.0333     0.8327
   price  int64     0.0%     240 82696  530595 284336.5750 66376.4447

Price Distribution

R2 by Model

RMSE Convergence

Training Pipeline

from scomp_link import ScompLinkPipeline, ScompArtifact
import pandas as pd

df = pd.read_csv("train.csv")
pipe = ScompLinkPipeline("House Price Prediction")
pipe.import_and_clean_data(df)
pipe.select_variables(target_col="price")
pipe.choose_model("numerical_prediction")
results = pipe.run_pipeline(task_type="regression")

artifact = ScompArtifact()
artifact.set_model(pipe.model)
artifact.set_metrics(results["metrics"])
artifact.save("model.scomp")

▶ Output

R2: 0.8392

Ridge to GBR + Optuna

Performance Comparison

Config Update

Validate

scomp-link validate --artifact model.scomp --data test.csv --target price

▶ Output

R2: 0.8392 | RMSE: 24156 | MAE: 18234 | MAPE: 6.41%
Model performs within acceptable thresholds.

Residual Distribution

Feature Importance

SHAP Code

from scomp_link import ShapExplainer
explainer = ShapExplainer(model, X_train[:100])
explainer.explain(X_test)
fig = explainer.plot_importance()

Deploy

scomp-link serve --artifact model.scomp --port 8080
scomp-link export --artifact model.scomp --format onnx

▶ Output

Serving on http://localhost:8080/predict

Pipeline DSL (>> operator)

from scomp_link import CodeStep, DiffStep, SectionStep, SaveStep
from scomp_link.utils.report_html import ScompLinkHTMLReport

report = ScompLinkHTMLReport("Report")
(
    SectionStep("Code") >> CodeStep("x=1", "python", "Ex")
    >> DiffStep("old", "new", "python", "Diff")
    >> SaveStep("out.html")
).run(report)