import sys
sys.path.insert(0, "..")

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

from sklearn.cluster import KMeans, AgglomerativeClustering
from sklearn.preprocessing import StandardScaler
from sklearn.metrics import silhouette_score, silhouette_samples, adjusted_rand_score
from scipy.cluster.hierarchy import linkage, dendrogram, fcluster

from common.datasets import segmentasi_pelanggan

X7 = np.array([[1, 1], [1.5, 2], [3, 4], [5, 7], [3.5, 5], [4.5, 5], [3.5, 4.5]])
nama = list("ABCDEFG")
C = X7[[0, 2]].copy()   # centroid awal = A dan C

for it in range(1, 4):
    D = ((X7[:, None] - C[None]) ** 2).sum(axis=2)
    label = D.argmin(axis=1)
    print(f"Iterasi {it}: klaster = {dict(zip(nama, label.tolist()))}")
    C_baru = np.array([X7[label == j].mean(axis=0) for j in range(2)])
    sse = ((X7 - C_baru[label]) ** 2).sum()
    print(f"   centroid baru = {C_baru.round(3).tolist()}  SSE = {sse:.3f}")
    if np.allclose(C_baru, C):
        print("   konvergen!")
        break
    C = C_baru

plt.scatter(X7[:, 0], X7[:, 1], c=label, cmap="coolwarm", s=80)
plt.scatter(C[:, 0], C[:, 1], marker="X", s=200, c="k")
for n, (x, y) in zip(nama, X7):
    plt.annotate(n, (x + 0.08, y + 0.08))
plt.title("Hasil K-Means (k=2)"); plt.show()

df = segmentasi_pelanggan()
fitur = ["recency_hari", "frekuensi", "monetary_ribu"]
df[fitur].describe().round(1)

Xs = StandardScaler().fit_transform(df[fitur])

ks = range(1, 10)
sse = [KMeans(k, n_init=10, random_state=0).fit(Xs).inertia_ for k in ks]
sil = [silhouette_score(Xs, KMeans(k, n_init=10, random_state=0).fit_predict(Xs)) for k in ks if k > 1]

fig, ax = plt.subplots(1, 2, figsize=(11, 3.5))
ax[0].plot(list(ks), sse, "o-"); ax[0].set_title("Elbow (SSE vs k)"); ax[0].set_xlabel("k")
ax[1].plot(list(ks)[1:], sil, "s-", color="C1"); ax[1].set_title("Silhouette vs k"); ax[1].set_xlabel("k")
plt.tight_layout(); plt.show()
print("k terbaik (silhouette):", list(ks)[1:][int(np.argmax(sil))])

km = KMeans(4, n_init=10, random_state=0).fit(Xs)
s = silhouette_samples(Xs, km.labels_)
y0 = 0
for j in range(4):
    sj = np.sort(s[km.labels_ == j])
    plt.fill_betweenx(np.arange(y0, y0 + len(sj)), 0, sj, alpha=0.7, label=f"klaster {j}")
    y0 += len(sj) + 10
plt.axvline(s.mean(), color="k", ls="--", label=f"rata-rata {s.mean():.2f}")
plt.xlabel("silhouette"); plt.legend(); plt.title("Silhouette per titik (k=4)"); plt.show()

df["klaster"] = km.labels_
profil = df.groupby("klaster")[fitur].mean().round(1)
profil["jumlah"] = df["klaster"].value_counts().sort_index()
profil["porsi_pendapatan_%"] = (df.groupby("klaster")["monetary_ribu"].sum() / df["monetary_ribu"].sum() * 100).round(1)
profil

# Karena data sintetis, kita bisa cek terhadap segmen asli (di dunia nyata label ini tidak ada!)
print("Adjusted Rand Index vs segmen asli:", round(adjusted_rand_score(df["segmen_asli"], km.labels_), 3))

rng = np.random.default_rng(0)
A = np.vstack([rng.normal([0, 0], 0.5, (30, 2)), rng.normal([4, 4], 0.5, (30, 2)), [[25, 25], [26, 24]]])

def k_medoids(X, medoid, iters=50):
    D = np.sqrt(((X[:, None] - X[None]) ** 2).sum(axis=2))
    for _ in range(iters):
        lab = D[:, medoid].argmin(axis=1)
        baru = [int(np.where(lab == j)[0][D[np.ix_(lab == j, lab == j)].sum(axis=1).argmin()]) for j in range(len(medoid))]
        if baru == medoid:
            break
        medoid = baru
    return lab, X[medoid]

km2 = KMeans(2, init=A[[0, 30]], n_init=1).fit(A)
lab_m, pusat_m = k_medoids(A, [0, 30])

fig, ax = plt.subplots(1, 2, figsize=(11, 4))
ax[0].scatter(A[:, 0], A[:, 1], c=km2.labels_, cmap="coolwarm"); ax[0].scatter(*km2.cluster_centers_.T, marker="X", s=200, c="k")
ax[0].set_title("K-Means: centroid tertarik outlier")
ax[1].scatter(A[:, 0], A[:, 1], c=lab_m, cmap="coolwarm"); ax[1].scatter(*pusat_m.T, marker="X", s=200, c="k")
ax[1].set_title("K-Medoids: medoid = titik nyata")
plt.show()
print("centroid K-Means :", km2.cluster_centers_.round(2).tolist())
print("medoid           :", pusat_m.round(2).tolist())

sampel = Xs[rng.choice(len(Xs), 40, replace=False)]
fig, ax = plt.subplots(1, 3, figsize=(15, 4))
for a, metode in zip(ax, ["single", "complete", "average"]):
    Z = linkage(sampel, method=metode)
    dendrogram(Z, ax=a, no_labels=True)
    a.set_title(f"{metode} linkage")
plt.tight_layout(); plt.show()

for metode in ["single", "complete", "average", "ward"]:
    lab = AgglomerativeClustering(n_clusters=4, linkage=metode).fit_predict(Xs)
    print(f"{metode:9s} silhouette={silhouette_score(Xs, lab):.3f}  ARI vs asli={adjusted_rand_score(df['segmen_asli'], lab):.3f}  ukuran={np.bincount(lab).tolist()}")

# 1.

# 2.

# 3.