emailspamclassifier / src /mlflow_tracking.py
vettri06's picture
Add inference source modules
e40ee72 verified
Raw History Blame Contribute Delete
4.5 kB
"""
MLflow experiment tracking module:
Handles run initialization, hyperparameter logging, multi-class evaluation metrics,
confusion matrix artifacts, and model pipeline tracking.
"""
import os
import mlflow
import mlflow.sklearn
from typing import Dict, Any
EXPERIMENT_NAME = "Email_Threat_Classification_SOC"
# Environment configuration
os.environ["MLFLOW_ALLOW_FILE_STORE"] = "true"
os.environ["MLFLOW_DISABLE_AGENT_HINT"] = "1"
def setup_mlflow(tracking_dir: str = None) -> str:
"""
Configures MLflow tracking with an SQLite database backend and artifact store.
"""
if tracking_dir is None:
tracking_dir = os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "mlruns"))
os.makedirs(tracking_dir, exist_ok=True)
db_file = os.path.join(tracking_dir, "mlflow.db")
db_path = db_file.replace("\\", "/")
uri = f"sqlite:///{db_path}"
# Ensure existing experiment and run artifact locations match the local environment
if os.path.exists(db_file):
try:
import sqlite3
conn = sqlite3.connect(db_file)
cur = conn.cursor()
current_exp1_dir = os.path.join(tracking_dir, "1").replace("\\", "/")
current_exp0_dir = os.path.join(tracking_dir, "0").replace("\\", "/")
cur.execute("UPDATE experiments SET artifact_location = ? WHERE experiment_id IN ('1', 1)", (f"file:{current_exp1_dir}",))
cur.execute("UPDATE experiments SET artifact_location = ? WHERE experiment_id IN ('0', 0)", (f"file:{current_exp0_dir}",))
for row in cur.execute("SELECT run_uuid, artifact_uri FROM runs").fetchall():
r_uuid, art_uri = row
if art_uri and "mlruns/1/" in art_uri:
new_uri = f"file:{current_exp1_dir}/{r_uuid}/artifacts"
cur.execute("UPDATE runs SET artifact_uri = ? WHERE run_uuid = ?", (new_uri, r_uuid))
conn.commit()
conn.close()
except Exception:
pass
mlflow.set_tracking_uri(uri)
mlflow.set_experiment(EXPERIMENT_NAME)
return uri
def log_experiment_run(
run_name: str,
model_type: str,
params: Dict[str, Any],
metrics: Dict[str, float],
artifacts: Dict[str, str] = None,
pipeline=None,
tags: Dict[str, str] = None
) -> str:
"""
Logs an atomic experiment run with parameters, metrics, figures, and models.
"""
with mlflow.start_run(run_name=run_name) as run:
# Standard cyber defense tags
default_tags = {
"model_type": model_type,
"domain": "Cybersecurity / Email Threat Intelligence",
"framework": "scikit-learn / xgboost",
"classes": "Legitimate, Spam, Suspicious"
}
if tags:
default_tags.update(tags)
mlflow.set_tags(default_tags)
# Log parameters
flat_params = {}
for k, v in params.items():
flat_params[k] = str(v)
flat_params["model_algorithm"] = model_type
mlflow.log_params(flat_params)
# Log metrics
for metric_name, metric_val in metrics.items():
if isinstance(metric_val, (int, float)):
mlflow.log_metric(metric_name, float(metric_val))
# Log artifact files (confusion matrix images, reports)
if artifacts:
for art_name, art_path in artifacts.items():
if os.path.exists(art_path):
mlflow.log_artifact(art_path, artifact_path="evaluation_artifacts")
# Log model pipeline artifact cleanly
if pipeline is not None:
try:
import tempfile
import joblib
with tempfile.TemporaryDirectory() as tmp_dir:
model_path = os.path.join(tmp_dir, "model_pipeline.joblib")
joblib.dump(pipeline, model_path)
mlflow.log_artifact(model_path, artifact_path="model_pipeline")
except Exception as e:
print(f"[Notice] Model artifact logging notice: {e}")
run_id = run.info.run_id
return run_id
def get_experiment_summary() -> list:
"""
Retrieves summary table of all logged runs in the experiment.
"""
experiment = mlflow.get_experiment_by_name(EXPERIMENT_NAME)
if not experiment:
return []
runs = mlflow.search_runs(experiment_ids=[experiment.experiment_id])
return runs.to_dict(orient="records")