Spaces:
Sleeping
Sleeping
File size: 5,880 Bytes
bf68d2b 5fbb6c5 bf68d2b | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 | """Train and persist the dual-task obesity models.
Two head-to-head model comparisons are run on the UCI Obesity Levels
dataset (`aiml2021/obesity`):
- Regression head — predict BMI from demographics + habits + activity.
Ridge baseline vs XGBRegressor.
- Classification head — predict the 7-class obesity level (NObeyesdad).
LogisticRegression baseline vs XGBClassifier.
Whichever model wins on the held-out test fold is persisted, together
with feature columns, baseline metrics, and per-class breakdown in
``models/numeric_metadata.json``.
"""
from __future__ import annotations
import json
from pathlib import Path
import joblib
from sklearn.linear_model import LogisticRegression, Ridge
from sklearn.metrics import (
accuracy_score,
classification_report,
f1_score,
mean_absolute_error,
r2_score,
)
from sklearn.model_selection import train_test_split
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import LabelEncoder, StandardScaler
from xgboost import XGBClassifier, XGBRegressor
from .obesity import OBESITY_LEVELS, build_features, load
MODELS_DIR = Path(__file__).resolve().parents[2] / "models"
SEED = 42
def train_regressor(X_train, X_test, y_train, y_test):
ridge = Pipeline([
("scaler", StandardScaler()),
("model", Ridge(alpha=1.0)),
]).fit(X_train, y_train)
ridge_mae = mean_absolute_error(y_test, ridge.predict(X_test))
ridge_r2 = r2_score(y_test, ridge.predict(X_test))
xgb = XGBRegressor(
n_estimators=400, max_depth=5, learning_rate=0.05,
subsample=0.9, colsample_bytree=0.9, random_state=SEED,
).fit(X_train, y_train)
xgb_mae = mean_absolute_error(y_test, xgb.predict(X_test))
xgb_r2 = r2_score(y_test, xgb.predict(X_test))
if xgb_mae <= ridge_mae:
return xgb, "XGBRegressor", {"mae": xgb_mae, "r2": xgb_r2}, {"ridge_mae": ridge_mae, "ridge_r2": ridge_r2}
return ridge, "Ridge", {"mae": ridge_mae, "r2": ridge_r2}, {"xgb_mae": xgb_mae, "xgb_r2": xgb_r2}
def train_classifier(X_train, X_test, y_train, y_test):
logit = Pipeline([
("scaler", StandardScaler()),
("model", LogisticRegression(max_iter=2000)),
]).fit(X_train, y_train)
logit_pred = logit.predict(X_test)
logit_acc = accuracy_score(y_test, logit_pred)
logit_f1 = f1_score(y_test, logit_pred, average="macro")
xgb = XGBClassifier(
n_estimators=400, max_depth=5, learning_rate=0.05,
subsample=0.9, colsample_bytree=0.9, random_state=SEED,
eval_metric="mlogloss", num_class=len(OBESITY_LEVELS),
).fit(X_train, y_train)
xgb_pred = xgb.predict(X_test)
xgb_acc = accuracy_score(y_test, xgb_pred)
xgb_f1 = f1_score(y_test, xgb_pred, average="macro")
if xgb_f1 >= logit_f1:
return (
xgb, "XGBClassifier",
{"accuracy": xgb_acc, "macro_f1": xgb_f1},
{"logit_accuracy": logit_acc, "logit_macro_f1": logit_f1},
xgb_pred,
)
return (
logit, "LogisticRegression",
{"accuracy": logit_acc, "macro_f1": logit_f1},
{"xgb_accuracy": xgb_acc, "xgb_macro_f1": xgb_f1},
logit_pred,
)
def main() -> None:
print("Loading UCI Obesity Levels dataset...")
df = load()
ds = build_features(df)
X = ds.features.astype("float64")
y_bmi = ds.bmi.values
label_enc = LabelEncoder().fit(OBESITY_LEVELS)
y_cls = label_enc.transform(ds.label.values)
X_train, X_test, y_bmi_train, y_bmi_test, y_cls_train, y_cls_test = train_test_split(
X, y_bmi, y_cls, test_size=0.2, random_state=SEED, stratify=y_cls,
)
print("Training regressor (Ridge vs XGB)...")
reg, reg_name, reg_metrics, reg_baseline = train_regressor(X_train, X_test, y_bmi_train, y_bmi_test)
print(f" -> chose {reg_name}: {reg_metrics}")
print(f" baseline: {reg_baseline}")
print("Training classifier (LogisticRegression vs XGB)...")
clf, clf_name, clf_metrics, clf_baseline, clf_pred = train_classifier(
X_train, X_test, y_cls_train, y_cls_test,
)
print(f" -> chose {clf_name}: {clf_metrics}")
print(f" baseline: {clf_baseline}")
MODELS_DIR.mkdir(parents=True, exist_ok=True)
joblib.dump(reg, MODELS_DIR / "numeric_regressor.pkl")
joblib.dump(clf, MODELS_DIR / "numeric_classifier.pkl")
joblib.dump(label_enc, MODELS_DIR / "numeric_label_encoder.pkl")
report = classification_report(
y_cls_test, clf_pred,
labels=list(range(len(OBESITY_LEVELS))),
target_names=OBESITY_LEVELS, output_dict=True, zero_division=0,
)
metadata = {
"dataset": "aiml2021/obesity",
"feature_columns": ds.feature_columns,
"classes": OBESITY_LEVELS,
"n_train": int(len(X_train)),
"n_test": int(len(X_test)),
"regressor": {
"name": reg_name,
"target": "BMI",
"metrics": {k: float(v) for k, v in reg_metrics.items()},
"baseline_metrics": {k: float(v) for k, v in reg_baseline.items()},
},
"classifier": {
"name": clf_name,
"target": "NObeyesdad",
"metrics": {k: float(v) for k, v in clf_metrics.items()},
"baseline_metrics": {k: float(v) for k, v in clf_baseline.items()},
"per_class": {
cls: {
"precision": float(report[cls]["precision"]),
"recall": float(report[cls]["recall"]),
"f1": float(report[cls]["f1-score"]),
"support": int(report[cls]["support"]),
}
for cls in OBESITY_LEVELS if cls in report
},
},
}
(MODELS_DIR / "numeric_metadata.json").write_text(json.dumps(metadata, indent=2))
print(f"\nSaved artifacts to {MODELS_DIR}")
if __name__ == "__main__":
main()
|