Confusion matrix, precision, recall, F1

import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.pipeline import Pipeline
from sklearn.compose import ColumnTransformer
from sklearn.preprocessing import StandardScaler, OneHotEncoder
from sklearn.impute import SimpleImputer
from sklearn.linear_model import LogisticRegression

df = pd.read_csv(DATA / "uy_ijara.csv")

son = ["xonalar", "maydon_m2", "qavat", "qavatlar_soni", "metro_km", "mebel", "qurilgan_yil"]
matn = ["tuman"]
X = df[son + matn]
y = df["ijaraga_berildi_30k"]

X_tr, X_te, y_tr, y_te = train_test_split(X, y, test_size=0.3, random_state=42, stratify=y)

quvur = Pipeline([
    ("tayyorla", ColumnTransformer([
        ("son", Pipeline([("toldir", SimpleImputer(strategy="median")),
                          ("shkala", StandardScaler())]), son),
        ("matn", OneHotEncoder(handle_unknown="ignore"), matn),
    ])),
    ("model", LogisticRegression(max_iter=2000)),
]).fit(X_tr, y_tr)

bashorat = quvur.predict(X_te)
print(f"test to'plami: {len(y_te)} kvartira")
print(f"accuracy     : {(bashorat == y_te).mean():.4f}")

from sklearn.metrics import confusion_matrix

cm = confusion_matrix(y_te, bashorat)
print("Confusion Matrix:")
print(cm)
print()
tn, fp, fn, tp = cm.ravel()
print(f"TN={tn}  FP={fp}  FN={fn}  TP={tp}")

show(pd.DataFrame([
    {"Belgi": "TN", "Soni": tn, "Model nima dedi": "sekin ketadi", "Haqiqatda": "sekin ketdi",
     "Biznes uchun": "To'g'ri. Reklama sarflanmadi, zarar yo'q"},
    {"Belgi": "FP", "Soni": fp, "Model nima dedi": "TEZ ketadi", "Haqiqatda": "sekin ketdi",
     "Biznes uchun": "Reklama puli YONDI — bo'sh e'longa sarflandi"},
    {"Belgi": "FN", "Soni": fn, "Model nima dedi": "sekin ketadi", "Haqiqatda": "TEZ ketdi",
     "Biznes uchun": "Imkoniyat QO'LDAN KETDI — e'tibor bermadik"},
    {"Belgi": "TP", "Soni": tp, "Model nima dedi": "TEZ ketadi", "Haqiqatda": "tez ketdi",
     "Biznes uchun": "To'g'ri. Reklama o'z joyiga tushdi"},
]))

show(pd.DataFrame([
    {"Metrika": "Precision", "Formula": "TP / (TP + FP)",
     "Savoli": "Model 'ha' deganda, qanchasida to'g'ri chiqdi?",
     "Kim so'raydi": "Pul sarflaydigan odam"},
    {"Metrika": "Recall", "Formula": "TP / (TP + FN)",
     "Savoli": "Bor 'ha' larning qanchasini topdi?",
     "Kim so'raydi": "Imkoniyatni qo'ldan chiqarmaslik kerak bo'lgan odam"},
]))

from sklearn.metrics import precision_score, recall_score, f1_score, accuracy_score

p = precision_score(y_te, bashorat)
r = recall_score(y_te, bashorat)
f1 = f1_score(y_te, bashorat)

print(f"Accuracy : {accuracy_score(y_te, bashorat):.4f}  = (TP+TN)/jami = ({tp}+{tn})/{len(y_te)}")
print(f"Precision: {p:.4f}  = TP/(TP+FP) = {tp}/({tp}+{fp})")
print(f"Recall   : {r:.4f}  = TP/(TP+FN) = {tp}/({tp}+{fn})")
print(f"F1       : {f1:.4f}")

hollar = [
    ("Muvozanatli model", 0.75, 0.80),
    ("Dangasa model (faqat 1 ta 'ha')", 1.00, 0.01),
    ("Saxiy model (hammaga 'ha')", 0.51, 1.00),
]
qatorlar = []
for nom, pr, rc in hollar:
    oddiy = (pr + rc) / 2
    garmonik = 2 * pr * rc / (pr + rc)
    qatorlar.append({"Model": nom, "Precision": pr, "Recall": rc,
                     "Oddiy o'rtacha": round(oddiy, 3), "F1 (garmonik)": round(garmonik, 3)})
show(pd.DataFrame(qatorlar))

print(f"Model nechta marta 'tez ketadi' dedi : {bashorat.sum()} ta")
print(f"Test to'plamida haqiqatan tez ketgani: {y_te.sum()} ta")
print()
print("Model saxiymi yoki qattiqmi:",
      "SAXIY" if bashorat.sum() > y_te.sum() else "QATTIQ")

from sklearn.metrics import classification_report

print(classification_report(y_te, bashorat,
                            target_names=["sekin ketdi", "tez ketdi"],
                            digits=4))

show(pd.DataFrame([
    {"Vaziyat": "Reklama budjeti cheklangan", "Ustun qo'yiladi": "Precision",
     "Nega": "Bo'sh e'longa pul sarflanmasin (FP qimmat)"},
    {"Vaziyat": "Har bir imkoniyat qimmat", "Ustun qo'yiladi": "Recall",
     "Nega": "Tez ketadigan kvartirani o'tkazib yubormaylik (FN qimmat)"},
    {"Vaziyat": "Kasallikni aniqlash", "Ustun qo'yiladi": "Recall",
     "Nega": "Kasalni sog'lom deb yuborish — eng xavfli xato"},
    {"Vaziyat": "Spam filtri", "Ustun qo'yiladi": "Precision",
     "Nega": "Muhim xatni spam ga tashlash — eng yomon xato"},
    {"Vaziyat": "Ikkisi ham muhim", "Ustun qo'yiladi": "F1",
     "Nega": "Muvozanat; bittasi past bo'lsa F1 ham past"},
    {"Vaziyat": "Sinflar juda nomutanosib", "Ustun qo'yiladi": "macro F1 / PR-AUC",
     "Nega": "Accuracy ham, weighted ham aldaydi (8-bob)"},
]))

import numpy as np

# Nomutanosib holat yasaymiz: "tez ketdi" ulushini ~5% ga tushiramiz
np.random.seed(42)
tez = y_te[y_te == 1].index
sekin = y_te[y_te == 0].index
tanlangan = list(sekin) + list(np.random.choice(tez, size=int(len(sekin) * 0.05), replace=False))

y_nom = y_te.loc[tanlangan]
print(f"Yangi to'plam: {len(y_nom)} qator, 'tez ketdi' ulushi {y_nom.mean():.1%}")
print()

dangasa = np.zeros(len(y_nom), dtype=int)   # hammaga "sekin ketadi" deydi
print(f"DANGASA model (hammaga 'sekin'):")
print(f"  Accuracy : {accuracy_score(y_nom, dangasa):.4f}   <-- chiroyli!")
print(f"  Recall   : {recall_score(y_nom, dangasa, zero_division=0):.4f}   <-- BITTA ham topmadi")
print(f"  F1       : {f1_score(y_nom, dangasa, zero_division=0):.4f}")

show(pd.DataFrame([
    {"Tushuncha": "Confusion matrix", "Bitta jumla bilan": "To'rtta katak: TN, FP, FN, TP — butun hikoya shunda"},
    {"Tushuncha": "FP va FN", "Bitta jumla bilan": "Ikki xil xato, ikki xil narx — accuracy ularni ajratmaydi"},
    {"Tushuncha": "Precision", "Bitta jumla bilan": "Model 'ha' deganda qanchasida to'g'ri — TP/(TP+FP)"},
    {"Tushuncha": "Recall", "Bitta jumla bilan": "Bor 'ha' larning qanchasini topdi — TP/(TP+FN)"},
    {"Tushuncha": "F1", "Bitta jumla bilan": "Garmonik o'rtacha; bittasi past bo'lsa F1 ham past"},
    {"Tushuncha": "macro vs weighted", "Bitta jumla bilan": "Nomutanosibda macro ni o'qi — weighted yashiradi"},
    {"Tushuncha": "Metrika tanlash", "Bitta jumla bilan": "Biznes qarori: FP qimmatmi yoki FN?"},
]))

show(pd.DataFrame([
    {"Metrika": "Accuracy", "Qiymat": f"{accuracy_score(y_te, bashorat):.4f}", "O'qilishi": "Umumiy to'g'rilik"},
    {"Metrika": "Precision", "Qiymat": f"{p:.4f}", "O'qilishi": f"'Ha' deganlarining {p:.0%} i to'g'ri"},
    {"Metrika": "Recall", "Qiymat": f"{r:.4f}", "O'qilishi": f"Imkoniyatlarning {r:.0%} ini topdi"},
    {"Metrika": "F1", "Qiymat": f"{f1:.4f}", "O'qilishi": "Muvozanat bahosi"},
    {"Metrika": "FP (yongan reklama)", "Qiymat": str(fp), "O'qilishi": "Shuncha bo'sh e'longa pul ketdi"},
    {"Metrika": "FN (qo'ldan ketgan)", "Qiymat": str(fn), "O'qilishi": "Shuncha imkoniyat sezilmadi"},
]))