Agent 1 β€” Structural Risk Random Forest

Random Forest that estimates defect risk from 21 NASA/PROMISE code metrics. Part of a three-agent DevSecOps auditor (metrics β†’ transformer β†’ gatekeeper).

Model details

Field Value
Version 1.0.0 (initial release)
Artifact model.joblib (bundle: model, raw_cols, feature_cols, threshold)
Task Binary tabular classification (defect / no defect)
Threshold 0.3183 (tuned on training CV with F2)
Training data NASA/PROMISE PC1 (pc1.csv)
Dataset version Not pinned β€” static PC1 snapshot, no revision hash recorded
Seed 42

metadata.json mirrors the dataset, threshold, and best params.

Input features

21 numeric metrics:

loc, vg, evg, ivg, n, v, l, d, i, e, b, t,
locode, locomment, loccodeandcomment, loblank,
uniq_op, uniq_opnd, total_op, total_opnd, branchcount

Five derived features are added before prediction. Final feature_cols order (26 columns, must match exactly):

loc, vg, evg, ivg, n, v, l, d, i, e, b, t,
locode, locomment, loccodeandcomment, loblank,
uniq_op, uniq_opnd, total_op, total_opnd, branchcount,
complexity_density, effort_per_loc, comment_ratio, branch_density, operand_reuse

Preprocessing

  • All 21 inputs must be numeric; no missing values (nothing is imputed).
  • Derive ratios with +1 denominators:
complexity_density = vg / (loc + 1)
effort_per_loc     = e / (loc + 1)
comment_ratio      = locomment / (loc + 1)
branch_density     = branchcount / (loc + 1)
operand_reuse      = total_opnd / (uniq_opnd + 1)
  • Select b["feature_cols"] in the order above.

Output

Raw: predict_proba(X)[0, 1] β€” defect risk in [0, 1], uncalibrated. high_risk = score >= 0.3183.

Reference wrapper output:

{"struct_risk_score": 0.5105, "high_risk": true, "threshold": 0.3183, "status": "ok", "reason": ""}

Usage

import joblib
import pandas as pd

b = joblib.load("model.joblib")
df = pd.DataFrame([metrics])[b["raw_cols"]].apply(pd.to_numeric)

df["complexity_density"] = df["vg"] / (df["loc"] + 1)
df["effort_per_loc"]     = df["e"] / (df["loc"] + 1)
df["comment_ratio"]      = df["locomment"] / (df["loc"] + 1)
df["branch_density"]     = df["branchcount"] / (df["loc"] + 1)
df["operand_reuse"]      = df["total_opnd"] / (df["uniq_opnd"] + 1)

X = df[b["feature_cols"]]
score = b["model"].predict_proba(X)[0, 1]
high_risk = score >= b["threshold"]

Trained on NASA PROMISE PC1. Metrics from other languages or domains are out-of-distribution.

License

Not specified. Update the license field before publishing.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support