import matplotlib.pyplot as plt

num_cols = df.select_dtypes("number").columns.tolist()
n = len(num_cols)
fig, axes = plt.subplots(2, 4, figsize=(14, 5))

for ax, col in zip(axes.flat, num_cols):
    df[col].dropna().hist(ax=ax, bins=30, color="#2563eb", edgecolor="none", alpha=.85)
    ax.set_title(col, fontsize=9)
    ax.set_xlabel("")

fig.tight_layout()
plt.show()

def iqr_outliers(series, k=1.5):
    """Return boolean mask — True where value is outside k*IQR fences."""
    q1, q3 = series.quantile([0.25, 0.75])
    iqr = q3 - q1
    return (series < q1 - k*iqr) | (series > q3 + k*iqr)

for col in num_cols:
    mask = iqr_outliers(df[col].dropna())
    print(f"{col:<20} {mask.sum():>4} outliers  ({mask.mean()*100:.1f}%)")

missing = (
    df.isnull()
    .sum()
    .sort_values(ascending=False)
)
missing_pct = missing / len(df) * 100

pd.DataFrame({
    "missing_n"  : missing,
    "missing_pct": missing_pct.round(1)
}[missing > 0]

import seaborn as sns

# Heatmap — white = present, blue = missing
fig, ax = plt.subplots(figsize=(10, 4))
sns.heatmap(
    df.isnull().T,
    cbar=False,
    cmap="Blues",
    yticklabels=True,
    ax=ax
)
ax.set_title("Missingness heatmap (blue = missing)")
plt.show()

# Test for MCAR with Little's test (requires missingno or pingouin)
from scipy import stats
# Quick proxy: correlation between missing indicator and other columns
for col in ["insulin", "bmi"]:
    indicator = df[col].isnull().astype(int)
    r, p = stats.pointbiserialr(indicator, df["outcome"])
    print(f"{col} missing ~ outcome:  r={r:.3f}, p={p:.4f}")

from sklearn.impute import SimpleImputer, KNNImputer

# MCAR / low % → median imputation (robust to outliers)
median_imp = SimpleImputer(strategy="median")
df[["bmi", "glucose"]] = median_imp.fit_transform(df[["bmi", "glucose"]])

# MAR → KNN imputation (uses other columns as context)
knn_imp = KNNImputer(n_neighbors=5)
df[["insulin", "skin_thickness"]] = knn_imp.fit_transform(
    df[["insulin", "skin_thickness", "age", "bmi"]]
)[:, :2]

# Confirm: no nulls remain
print("Remaining nulls:", df.isnull().sum().sum())

import matplotlib.pyplot as plt

# Pearson correlation of every feature with the target
target = "outcome"
corrs = df.drop(columns=[target]).corrwith(df[target]).abs().sort_values(ascending=False)

corrs.plot(kind="bar", color=["#ef4444" if v > 0.6 else "#2563eb" for v in corrs])
plt.axhline(0.6, color="#ef4444", linestyle="--", label="leakage threshold")
plt.title("Feature–target correlations (|r|)")
plt.tight_layout(); plt.show()

from sklearn.pipeline      import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.impute        import SimpleImputer
from sklearn.ensemble      import RandomForestClassifier
from sklearn.model_selection import cross_val_score, train_test_split

X = df.drop(columns=["outcome"])
y = df["outcome"]

# Split FIRST — nothing touches X_test until evaluation
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, random_state=42, stratify=y
)

# Pipeline: impute → scale → model
# Imputer and scaler are fit ONLY on training folds
pipe = Pipeline([
    ("imputer", SimpleImputer(strategy="median")),
    ("scaler",  StandardScaler()),
    ("model",   RandomForestClassifier(n_estimators=100, random_state=42)),
])

cv_scores = cross_val_score(pipe, X_train, y_train, cv=5, scoring="roc_auc")
print(f"CV AUC: {cv_scores.mean():.3f} ± {cv_scores.std():.3f}")

import pandas as pd; import numpy as np
import matplotlib.pyplot as plt; import seaborn as sns

df = pd.read_csv("your_data.csv")

# ── Q1. What is the shape? ──────────────────────────────
print("Shape:", df.shape)

# ── Q2. What are the types? ─────────────────────────────
df.info()

# ── Q3. Do the ranges make sense? ──────────────────────
print(df.describe().round(2))

# ── Q4. Are there impossible zeros? ────────────────────
print((df == 0).sum()[(df == 0).sum() > 0])

# ── Q5. How much is missing? ────────────────────────────
miss = df.isnull().mean().sort_values(ascending=False)
print(miss[miss > 0])

# ── Q6. What is the class balance? ─────────────────────
print(df["target"].value_counts(normalize=True))

# ── Q7. Are there suspicious feature–target correlations? ─
high_corr = df.corrwith(df["target"]).abs()
print(high_corr[high_corr > 0.6])

# ── Q8. Are any features duplicates or near-constants? ──
print("Near-zero variance:", df.std()[df.std() < 0.01].index.tolist())
print("Duplicate cols:",   df.T.duplicated().sum())