Delete README.md
Browse files
README.md
DELETED
|
@@ -1,79 +0,0 @@
|
|
| 1 |
-
# Agent 1 — Structural Risk (Random Forest)
|
| 2 |
-
|
| 3 |
-
Predicts defect risk from NASA/PROMISE code metrics. Trained on `data/pc1.csv`.
|
| 4 |
-
This model is a single artifact, so its README lives beside it as
|
| 5 |
-
`agent1_rf.README.md`.
|
| 6 |
-
|
| 7 |
-
## Files
|
| 8 |
-
|
| 9 |
-
| File | Contents |
|
| 10 |
-
|---|---|
|
| 11 |
-
| `agent1_rf.joblib` | Model bundle: `model`, `raw_cols`, `feature_cols`, `threshold` |
|
| 12 |
-
| `agent1_rf.meta.json` | Human-readable copy: dataset, seed, threshold, best params |
|
| 13 |
-
|
| 14 |
-
Code: `agents/agent1_metrics.py`.
|
| 15 |
-
|
| 16 |
-
## Input
|
| 17 |
-
|
| 18 |
-
One dict containing all 21 metrics:
|
| 19 |
-
|
| 20 |
-
```
|
| 21 |
-
loc, vg, evg, ivg, n, v, l, d, i, e, b, t, locode, locomment,
|
| 22 |
-
loccodeandcomment, loblank, uniq_op, uniq_opnd, total_op, total_opnd, branchcount
|
| 23 |
-
```
|
| 24 |
-
|
| 25 |
-
Keys are normalized (lowercased, stripped, parentheses removed), so `v(g)` and
|
| 26 |
-
`iv(G)` become `vg` and `ivg`. Values must be numeric. Five derived ratios
|
| 27 |
-
are computed internally (`complexity_density`, `effort_per_loc`, `comment_ratio`,
|
| 28 |
-
`branch_density`, `operand_reuse`) — do not supply them.
|
| 29 |
-
|
| 30 |
-
Missing keys or non-numeric values return an error; nothing is imputed at
|
| 31 |
-
inference time.
|
| 32 |
-
|
| 33 |
-
## Output
|
| 34 |
-
|
| 35 |
-
Success:
|
| 36 |
-
```json
|
| 37 |
-
{
|
| 38 |
-
"struct_risk_score": 0.5105,
|
| 39 |
-
"high_risk": true,
|
| 40 |
-
"threshold": 0.3183,
|
| 41 |
-
"status": "ok",
|
| 42 |
-
"reason": ""
|
| 43 |
-
}
|
| 44 |
-
```
|
| 45 |
-
|
| 46 |
-
Error (note: no `threshold` key):
|
| 47 |
-
```json
|
| 48 |
-
{
|
| 49 |
-
"struct_risk_score": null,
|
| 50 |
-
"high_risk": null,
|
| 51 |
-
"status": "error",
|
| 52 |
-
"reason": "missing features: ['loc', 'vg']"
|
| 53 |
-
}
|
| 54 |
-
```
|
| 55 |
-
|
| 56 |
-
| Key | Success | Error |
|
| 57 |
-
|---|---|---|
|
| 58 |
-
| `struct_risk_score` | float 0–1 | `null` |
|
| 59 |
-
| `high_risk` | `score >= threshold` | `null` |
|
| 60 |
-
| `threshold` | `0.3183` | key absent |
|
| 61 |
-
| `status` | `"ok"` | `"error"` |
|
| 62 |
-
| `reason` | `""` | error detail |
|
| 63 |
-
|
| 64 |
-
`struct_risk_score` is an uncalibrated model score. `status != "ok"` means
|
| 65 |
-
Agent 3 will return `REVIEW`.
|
| 66 |
-
|
| 67 |
-
## Usage
|
| 68 |
-
|
| 69 |
-
```bash
|
| 70 |
-
# Train (needs data/pc1.csv)
|
| 71 |
-
python agents/agent1_metrics.py train --data data/pc1.csv
|
| 72 |
-
|
| 73 |
-
# Predict
|
| 74 |
-
python -c "from agents.agent1_metrics import predict_struct_risk; \
|
| 75 |
-
print(predict_struct_risk({'loc': 200, 'vg': 25, ...}))"
|
| 76 |
-
|
| 77 |
-
# Test
|
| 78 |
-
python test_agent1.py
|
| 79 |
-
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|