File size: 4,648 Bytes
aad0df8 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 | from __future__ import annotations
import json
import joblib
import numpy as np
import pandas as pd
import trackio
from sentence_transformers import SentenceTransformer
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import (
accuracy_score,
confusion_matrix,
f1_score,
precision_score,
recall_score,
)
from train import ARTIFACT_DIR, DATA_DIR
BASE_MODEL = "sentence-transformers/all-MiniLM-L6-v2"
OUTPUT_DIR = ARTIFACT_DIR.parent / "protocol-guardian-minilm-linear"
def load(name: str) -> tuple[list[str], np.ndarray]:
frame = pd.read_parquet(DATA_DIR / f"{name}.parquet")
return (
frame["text"].tolist(),
frame["label"].to_numpy(dtype=np.int64, copy=True),
)
def metrics(labels: np.ndarray, predictions: np.ndarray) -> dict:
return {
"accuracy": float(accuracy_score(labels, predictions)),
"precision": float(precision_score(labels, predictions, zero_division=0)),
"recall": float(recall_score(labels, predictions, zero_division=0)),
"f1": float(f1_score(labels, predictions, zero_division=0)),
"confusion_matrix": confusion_matrix(labels, predictions).tolist(),
"examples": len(labels),
}
def main() -> None:
train_text, train_labels = load("train")
validation_text, validation_labels = load("validation")
test_text, test_labels = load("test")
encoder = SentenceTransformer(BASE_MODEL, device="cpu")
embeddings = {
"train": encoder.encode(
train_text,
batch_size=128,
normalize_embeddings=True,
show_progress_bar=True,
),
"validation": encoder.encode(
validation_text,
batch_size=128,
normalize_embeddings=True,
show_progress_bar=True,
),
"test": encoder.encode(
test_text,
batch_size=128,
normalize_embeddings=True,
show_progress_bar=True,
),
}
trackio.init(
project="protocol-guardian",
name="minilm-embedding-linear-v1",
config={
"base_model": BASE_MODEL,
"embedding_dimensions": embeddings["train"].shape[1],
"train_examples": len(train_labels),
"validation_template_family": "held-out from train",
"test_template_family": "held-out from train and validation",
},
)
candidates = []
for regularization in [0.03, 0.1, 0.3, 1.0, 3.0, 10.0]:
classifier = LogisticRegression(
C=regularization,
class_weight="balanced",
max_iter=2000,
random_state=2026,
)
classifier.fit(embeddings["train"], train_labels)
predictions = classifier.predict(embeddings["validation"])
score = f1_score(validation_labels, predictions)
candidates.append((score, regularization, classifier))
trackio.log(
{
"regularization_c": regularization,
"validation_f1": score,
"validation_accuracy": accuracy_score(
validation_labels,
predictions,
),
}
)
_, best_c, classifier = max(candidates, key=lambda item: item[0])
validation_predictions = classifier.predict(embeddings["validation"])
test_predictions = classifier.predict(embeddings["test"])
results = {
"model": "Protocol Guardian MiniLM Linear",
"base_model": BASE_MODEL,
"embedding_dimensions": int(embeddings["train"].shape[1]),
"linear_head_parameters": int(classifier.coef_.size + classifier.intercept_.size),
"selected_c": best_c,
"validation": metrics(validation_labels, validation_predictions),
"test": metrics(test_labels, test_predictions),
"test_split": "1200 examples from a third, unseen template family",
}
trackio.log(
{
"test_accuracy": results["test"]["accuracy"],
"test_f1": results["test"]["f1"],
"test_recall": results["test"]["recall"],
}
)
trackio.finish()
OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
joblib.dump(
{
"classifier": classifier,
"base_model": BASE_MODEL,
"labels": {0: "ROUTINE", 1: "HAZARDOUS"},
},
OUTPUT_DIR / "classifier.joblib",
compress=3,
)
(OUTPUT_DIR / "training_summary.json").write_text(
json.dumps(results, indent=2),
encoding="utf-8",
)
print(json.dumps(results, indent=2))
if __name__ == "__main__":
main()
|