15.6-dars: Extra Trees va OOB tahlili

Random Forest bo'linishi:
  1. max_features ta belgi tasodifiy tanlanadi
  2. HAR BIRI uchun ENG YAXSHI chegara qidiriladi
  3. Eng yaxshi (belgi, chegara) juftligi olinadi

Extra Trees bo'linishi:
  1. max_features ta belgi tasodifiy tanlanadi
  2. HAR BIRI uchun BITTA TASODIFIY chegara olinadi
  3. Shular orasidan eng yaxshisi tanlanadi

Natija:
  + chegara qidiruvi yo'q -> TEZ
  + ko'proq tasodifiylik -> rho kichikroq
  - har daraxt kuchsizroq (bias oshadi)

                        Random Forest    Extra Trees
bootstrap                 True (ha)       False (yo'q!)
chegara tanlash           eng yaxshi      tasodifiy
o'qitish tezligi          o'rtacha        tez
bitta daraxt kuchi        yuqori          past
daraxtlar korrelyatsiyasi o'rtacha        past
odatda kerak n_estimators 300-500         500-1000
oob_score                 ishlaydi        bootstrap=True kerak

Extra Trees yaxshiroq:
  + shovqin KO'P bo'lsa (tasodifiy chegara shovqinga moslashmaydi)
  + belgilar ko'p va o'zaro o'xshash
  + tezlik muhim
  + juda katta ma'lumot

Random Forest yaxshiroq:
  + signal aniq va kuchli (aniq chegaralar muhim)
  + belgilar kam
  + OOB baho kerak

Amalda: ikkalasini sinab ko'ring - farq odatda 1-3%

# daraxtlar sonining ta'sirini OOB bilan kuzatish
for ne in [50, 100, 200, 400, 800]:
    o = RandomForestClassifier(n_estimators=ne, oob_score=True,
                               random_state=0, warm_start=False).fit(X, y)
    print(ne, o.oob_score_)

# warm_start bilan bosqichma-bosqich
o = RandomForestClassifier(warm_start=True, oob_score=True, random_state=0)
for ne in [50, 100, 200]:
    o.set_params(n_estimators=ne).fit(X, y)

oob_decision_function_ - har o'quv namunasi uchun OOB ehtimollik

Undan CV siz qurish mumkin:
  - ROC va PR egri chiziqlari 14.9-bob
  - kalibrlash diagrammasi 14.10-bob
  - xatolar tahlili: qaysi namunalar noto'g'ri 14.13-bob
  - chegarani tanlash

Ehtiyot: NaN bo'lishi mumkin (namuna hech bir daraxtda OOB bo'lmasa)
         n_estimators kichik bo'lganda ko'p uchraydi

RandomForest(bootstrap=False)  -> har daraxt butun ma'lumotni ko'radi
  xilma-xillik faqat max_features dan
  OOB ishlamaydi
  ba'zan biroz yaxshiroq (kichik ma'lumotda)

ExtraTrees(bootstrap=True)     -> OOB yoqiladi
  xilma-xillik ko'proq

max_samples=0.5  -> har daraxtga ma'lumotning yarmi (tezlik uchun)

from sklearn.ensemble import ExtraTreesClassifier, ExtraTreesRegressor

e = ExtraTreesClassifier(n_estimators=500, max_features="sqrt",
                         bootstrap=True, oob_score=True,   # OOB uchun kerak
                         random_state=0, n_jobs=-1).fit(X, y)

# warm_start bilan o'rganish egri chizig'i
o = RandomForestClassifier(warm_start=True, oob_score=True, random_state=0)
for ne in [50, 100, 200, 400]:
    o.set_params(n_estimators=ne).fit(X, y)
    print(ne, o.oob_score_)

o.oob_decision_function_          # NaN bo'lishi mumkin - tekshiring
RandomForestRegressor(max_samples=0.3)   # katta ma'lumotda tezlik
QOIDA: ikkalasini sinang · ET uchun bootstrap=True (OOB uchun zarur) ·
       NaN ni tekshiring

ET: chegara tasodifiy, bootstrap yo'q (standart)
rho kichikroq, bias kattaroq, o'qitish 2-5x tez
Shovqinli ma'lumotda ko'pincha ustun
OOB: warm_start bilan egri chiziq, oob_decision_function_ bilan tahlil

"""Ikki ansamblning xossalari (real numpy/sklearn)."""

import numpy as np
from sklearn.ensemble import ExtraTreesClassifier, RandomForestClassifier
from sklearn.metrics import roc_auc_score
from sklearn.model_selection import train_test_split


def yarat(seed: int = 6, n: int = 3000, p: int = 20, shovqin: float = 0.12):
    rng = np.random.default_rng(seed)
    X = rng.normal(0, 1, (n, p))
    qoida = (((X[:, 0] > 0.3) & (X[:, 1] < 0.2))
             | ((X[:, 2] > 0.4) & (X[:, 3] > 0.0))
             | (X[:, 4] < -1.1))
    y = qoida.astype(int)
    alm = rng.random(n) < shovqin
    y[alm] = 1 - y[alm]
    return X, y


def main() -> None:
    X, y = yarat()
    Xtr, Xte, ytr, yte = train_test_split(X, y, test_size=0.3, random_state=0,
                                          stratify=y)

    print("=== 1. Asosiy solishtirish ===")
    modellar = {
        "RandomForest": RandomForestClassifier(n_estimators=300, random_state=0,
                                               n_jobs=1),
        "ExtraTrees": ExtraTreesClassifier(n_estimators=300, random_state=0,
                                           n_jobs=1),
    }
    for nom, m in modellar.items():
        m.fit(Xtr, ytr)
        a = roc_auc_score(yte, m.predict_proba(Xte)[:, 1])
        barglar = np.mean([e.get_n_leaves() for e in m.estimators_])
        chuqurlik = np.mean([e.get_depth() for e in m.estimators_])
        print(f"  {nom:<14}: AUC {a:.4f}, o'rtacha {barglar:.0f} barg, "
              f"chuqurlik {chuqurlik:.1f}")

    print("\n=== 2. Bitta daraxt kuchi va ansambl kuchi ===")
    for nom, m in modellar.items():
        bitta = [roc_auc_score(yte, e.predict_proba(Xte)[:, 1])
                 for e in m.estimators_[:30]]
        ansambl = roc_auc_score(yte, m.predict_proba(Xte)[:, 1])
        print(f"  {nom:<14}: bitta daraxt {np.mean(bitta):.4f}, "
              f"ansambl {ansambl:.4f}, foyda {ansambl - np.mean(bitta):+.4f}")

    print("\n=== 3. Daraxtlar orasidagi kelishmovchilik ===")
    for nom, m in modellar.items():
        P = np.array([e.predict(Xte) for e in m.estimators_[:20]])
        juftlar = [(P[i] != P[j]).mean()
                   for i in range(20) for j in range(i + 1, 20)]
        print(f"  {nom:<14}: o'rtacha {np.mean(juftlar):.4f} namunada farq")

    print("\n=== 4. Shovqin darajasi bo'yicha ===")
    print(f"  {'shovqin':>8} {'RF AUC':>9} {'ET AUC':>9} {'farq':>9}")
    for sh in [0.0, 0.05, 0.15, 0.30]:
        Xn, yn = yarat(shovqin=sh)
        Xa, Xb, ya, yb = train_test_split(Xn, yn, test_size=0.3, random_state=0,
                                          stratify=yn)
        r = RandomForestClassifier(n_estimators=300, random_state=0,
                                   n_jobs=1).fit(Xa, ya)
        e = ExtraTreesClassifier(n_estimators=300, random_state=0,
                                 n_jobs=1).fit(Xa, ya)
        ar = roc_auc_score(yb, r.predict_proba(Xb)[:, 1])
        ae = roc_auc_score(yb, e.predict_proba(Xb)[:, 1])
        print(f"  {sh:>8.2f} {ar:>9.4f} {ae:>9.4f} {ae - ar:>+9.4f}")
    print("  ⭐ Shovqin ko'p bo'lsa Extra Trees ko'pincha ustun")


if __name__ == "__main__":
    main()

=== 1. Asosiy solishtirish ===
  RandomForest  : AUC 0.8891, o'rtacha 247 barg, chuqurlik 19.4
  ExtraTrees    : AUC 0.8579, o'rtacha 899 barg, chuqurlik 31.2

=== 2. Bitta daraxt kuchi va ansambl kuchi ===
  RandomForest  : bitta daraxt 0.7190, ansambl 0.8891, foyda +0.1702
  ExtraTrees    : bitta daraxt 0.6028, ansambl 0.8579, foyda +0.2551

=== 3. Daraxtlar orasidagi kelishmovchilik ===
  RandomForest  : o'rtacha 0.3214 namunada farq
  ExtraTrees    : o'rtacha 0.4480 namunada farq

=== 4. Shovqin darajasi bo'yicha ===
   shovqin    RF AUC    ET AUC      farq
      0.00    1.0000    0.9828   -0.0172
      0.05    0.9545    0.9353   -0.0191
      0.15    0.8477    0.8361   -0.0116
      0.30    0.6817    0.6491   -0.0326
  ⭐ Shovqin ko'p bo'lsa Extra Trees ko'pincha ustun

"""warm_start bilan daraxtlar sonining ta'sirini kuzatish (real numpy/sklearn)."""

import warnings

import numpy as np
from sklearn.ensemble import ExtraTreesClassifier, RandomForestClassifier
from sklearn.metrics import roc_auc_score
from sklearn.model_selection import train_test_split


def yarat(seed: int = 10, n: int = 4000, p: int = 18, shovqin: float = 0.13):
    rng = np.random.default_rng(seed)
    X = rng.normal(0, 1, (n, p))
    qoida = (((X[:, 0] > 0.25) & (X[:, 1] < 0.25))
             | ((X[:, 2] > 0.5) & (X[:, 3] > -0.1))
             | (X[:, 4] < -1.0))
    y = qoida.astype(int)
    alm = rng.random(n) < shovqin
    y[alm] = 1 - y[alm]
    return X, y


def main() -> None:
    # kichik n_estimators da OOB bahosi to'liq bo'lmaydi -
    # bu aynan ko'rsatmoqchi bo'lgan hodisamiz, ogohlantirishni bostiramiz
    warnings.simplefilter("ignore")
    X, y = yarat()
    Xtr, Xte, ytr, yte = train_test_split(X, y, test_size=0.3, random_state=0,
                                          stratify=y)

    print("=== 1. RF: warm_start bilan OOB egri chizig'i ===")
    o = RandomForestClassifier(warm_start=True, oob_score=True, random_state=0,
                               n_jobs=1)
    print(f"  {'n_est':>7} {'OOB':>9} {'test AUC':>10} {'NaN':>6}")
    for ne in [10, 25, 50, 100, 200, 400]:
        o.set_params(n_estimators=ne).fit(Xtr, ytr)
        a = roc_auc_score(yte, o.predict_proba(Xte)[:, 1])
        nanlar = int(np.isnan(o.oob_decision_function_).any(axis=1).sum())
        print(f"  {ne:>7} {o.oob_score_:>9.4f} {a:>10.4f} {nanlar:>6}")

    print("\n=== 2. ExtraTrees: bootstrap kerak ===")
    e1 = ExtraTreesClassifier(n_estimators=200, random_state=0,
                              n_jobs=1).fit(Xtr, ytr)
    print(f"  bootstrap standart qiymati: {e1.bootstrap}")
    e2 = ExtraTreesClassifier(n_estimators=200, bootstrap=True, oob_score=True,
                              random_state=0, n_jobs=1).fit(Xtr, ytr)
    print(f"  bootstrap=True bilan OOB: {e2.oob_score_:.4f}")
    print(f"  test AUC (bootstrap=False): "
          f"{roc_auc_score(yte, e1.predict_proba(Xte)[:, 1]):.4f}")
    print(f"  test AUC (bootstrap=True):  "
          f"{roc_auc_score(yte, e2.predict_proba(Xte)[:, 1]):.4f}")

    print("\n=== 3. RF: bootstrap=False ===")
    for bs in [True, False]:
        r = RandomForestClassifier(n_estimators=300, bootstrap=bs,
                                   random_state=0, n_jobs=1).fit(Xtr, ytr)
        a = roc_auc_score(yte, r.predict_proba(Xte)[:, 1])
        print(f"  bootstrap={str(bs):<5}: test AUC {a:.4f}")

    print("\n=== 4. max_samples ta'siri ===")
    print(f"  {'max_samples':>12} {'OOB':>9} {'test AUC':>10} "
          f"{'o_rt barglar':>14}")
    for ms in [0.2, 0.5, 0.8, None]:
        r = RandomForestClassifier(n_estimators=300, max_samples=ms,
                                   oob_score=True, random_state=0,
                                   n_jobs=1).fit(Xtr, ytr)
        a = roc_auc_score(yte, r.predict_proba(Xte)[:, 1])
        barglar = np.mean([t.get_n_leaves() for t in r.estimators_])
        nom = "None" if ms is None else str(ms)
        print(f"  {nom:>12} {r.oob_score_:>9.4f} {a:>10.4f} {barglar:>14.1f}")
    print("  ⭐ max_samples - sifat va tezlik muvozanati")


if __name__ == "__main__":
    main()

=== 1. RF: warm_start bilan OOB egri chizig'i ===
    n_est       OOB   test AUC    NaN
       10    0.7846     0.8687      0
       25    0.8389     0.8757      0
       50    0.8650     0.8743      0
      100    0.8750     0.8776      0
      200    0.8761     0.8781      0
      400    0.8768     0.8756      0

=== 2. ExtraTrees: bootstrap kerak ===
  bootstrap standart qiymati: False
  bootstrap=True bilan OOB: 0.8043
  test AUC (bootstrap=False): 0.8644
  test AUC (bootstrap=True):  0.8578

=== 3. RF: bootstrap=False ===
  bootstrap=True : test AUC 0.8781
  bootstrap=False: test AUC 0.8776

=== 4. max_samples ta'siri ===
   max_samples       OOB   test AUC   o_rt barglar
           0.2    0.8707     0.8742           97.3
           0.5    0.8754     0.8746          199.7
           0.8    0.8768     0.8762          276.0
          None    0.8764     0.8781          314.4
  ⭐ max_samples - sifat va tezlik muvozanati

"""oob_decision_function_ dan kalibrlash va xato tahlili (real numpy/sklearn)."""

import numpy as np
from sklearn.calibration import calibration_curve
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import (average_precision_score, brier_score_loss,
                             precision_recall_curve, roc_auc_score)
from sklearn.model_selection import train_test_split


def yarat(seed: int = 2, n: int = 5000, p: int = 12):
    rng = np.random.default_rng(seed)
    X = rng.normal(0, 1, (n, p))
    kuch = (-2.2 + 1.3 * (X[:, 0] > 0.5) + 1.1 * (X[:, 1] < -0.5)
            + 1.5 * ((X[:, 2] > 0) & (X[:, 3] > 0)) + 0.6 * X[:, 4])
    y = (rng.random(n) < 1 / (1 + np.exp(-kuch))).astype(int)
    return X, y


def main() -> None:
    X, y = yarat()
    Xtr, Xte, ytr, yte = train_test_split(X, y, test_size=0.3, random_state=0,
                                          stratify=y)

    o = RandomForestClassifier(n_estimators=500, min_samples_leaf=5,
                               oob_score=True, random_state=0,
                               n_jobs=1).fit(Xtr, ytr)
    oob = o.oob_decision_function_[:, 1]
    test = o.predict_proba(Xte)[:, 1]

    print("=== 1. OOB va test ko'rsatkichlari ===")
    print(f"  NaN qatorlar: {int(np.isnan(oob).sum())}")
    print(f"  {'ko_rsatkich':<16} {'OOB':>9} {'test':>9}")
    print(f"  {'ROC AUC':<16} {roc_auc_score(ytr, oob):>9.4f} "
          f"{roc_auc_score(yte, test):>9.4f}")
    print(f"  {'PR AUC':<16} {average_precision_score(ytr, oob):>9.4f} "
          f"{average_precision_score(yte, test):>9.4f}")
    print(f"  {'Brier':<16} {brier_score_loss(ytr, oob):>9.4f} "
          f"{brier_score_loss(yte, test):>9.4f}")

    print("\n=== 2. OOB bilan kalibrlash diagrammasi ===")
    haqiqiy, bashorat = calibration_curve(ytr, oob, n_bins=8, strategy="quantile")
    print(f"  {'bashorat':>10} {'haqiqiy':>10} {'farq':>9}")
    for b, h in zip(bashorat, haqiqiy):
        print(f"  {b:>10.4f} {h:>10.4f} {h - b:>+9.4f}")

    print("\n=== 3. OOB bilan chegara tanlash ===")
    p_, r_, ch_ = precision_recall_curve(ytr, oob)
    f1 = 2 * p_[:-1] * r_[:-1] / np.maximum(p_[:-1] + r_[:-1], 1e-12)
    eng = int(np.argmax(f1))
    chegara = float(ch_[eng])
    print(f"  OOB bo'yicha eng yaxshi chegara: {chegara:.4f}")
    print(f"  OOB da F1: {f1[eng]:.4f}")
    from sklearn.metrics import f1_score
    print(f"  0.5 chegarada test F1:      "
          f"{f1_score(yte, (test > 0.5).astype(int)):.4f}")
    print(f"  tanlangan chegarada test F1: "
          f"{f1_score(yte, (test > chegara).astype(int)):.4f}")

    print("\n=== 4. OOB bilan xatolar tahlili ===")
    xato = (oob > 0.5).astype(int) != ytr.to_numpy() if hasattr(ytr, "to_numpy") \
        else (oob > 0.5).astype(int) != ytr
    print(f"  xato ulushi: {xato.mean():.4f}")
    ishonch = np.abs(oob - 0.5)
    print(f"  {'ishonch':>14} {'namunalar':>11} {'xato %':>9}")
    for past, yuqori, nom in [(0.0, 0.1, "0.0-0.1"), (0.1, 0.25, "0.1-0.25"),
                              (0.25, 0.4, "0.25-0.4"), (0.4, 0.51, "0.4-0.5")]:
        m = (ishonch >= past) & (ishonch < yuqori)
        print(f"  {nom:>14} {int(m.sum()):>11} {xato[m].mean():>8.1%}")
    print("  ⭐ OOB to'liq diagnostika beradi - CV siz")


if __name__ == "__main__":
    main()

=== 1. OOB va test ko'rsatkichlari ===
  NaN qatorlar: 0
  ko_rsatkich            OOB      test
  ROC AUC             0.7678    0.7604
  PR AUC              0.5880    0.5862
  Brier               0.1694    0.1710

=== 2. OOB bilan kalibrlash diagrammasi ===
    bashorat    haqiqiy      farq
      0.0707     0.0685   -0.0022
      0.1234     0.0892   -0.0341
      0.1790     0.1461   -0.0328
      0.2405     0.2311   -0.0094
      0.3059     0.2952   -0.0108
      0.3872     0.4018   +0.0146
      0.4801     0.4622   -0.0178
      0.6213     0.6872   +0.0659

=== 3. OOB bilan chegara tanlash ===
  OOB bo'yicha eng yaxshi chegara: 0.3173
  OOB da F1: 0.5875
  0.5 chegarada test F1:      0.4318
  tanlangan chegarada test F1: 0.5855

=== 4. OOB bilan xatolar tahlili ===
  xato ulushi: 0.2500
         ishonch   namunalar    xato %
         0.0-0.1         787    42.9%
        0.1-0.25        1083    31.0%
        0.25-0.4        1180    14.2%
         0.4-0.5         450     7.3%
  ⭐ OOB to'liq diagnostika beradi - CV siz

"""RF, ET va bagging regressiyada (real numpy/sklearn)."""

import numpy as np
from sklearn.ensemble import (BaggingRegressor, ExtraTreesRegressor,
                              RandomForestRegressor)
from sklearn.metrics import mean_absolute_error, r2_score
from sklearn.model_selection import KFold, cross_val_score, train_test_split
from sklearn.tree import DecisionTreeRegressor


def yarat(seed: int = 5, n: int = 3000, p: int = 20, shovqin: float = 1.2):
    rng = np.random.default_rng(seed)
    X = rng.normal(0, 1, (n, p))
    f = (2.0 * np.sin(1.5 * X[:, 0]) + 1.5 * X[:, 1] * (X[:, 2] > 0)
         - 1.0 * X[:, 3] ** 2 + 0.8 * X[:, 4]
         + 0.3 * X[:, 5:10].sum(axis=1))
    return X, f + rng.normal(0, shovqin, n), f


def main() -> None:
    X, y, f = yarat()
    Xtr, Xte, ytr, yte = train_test_split(X, y, test_size=0.3, random_state=0)
    cv = KFold(5, shuffle=True, random_state=0)

    print("=== 1. Uch ansambl ===")
    modellar = {
        "bagging": BaggingRegressor(DecisionTreeRegressor(random_state=0),
                                    n_estimators=300, random_state=0, n_jobs=1),
        "RandomForest": RandomForestRegressor(n_estimators=300,
                                              max_features=0.3,
                                              random_state=0, n_jobs=1),
        "ExtraTrees": ExtraTreesRegressor(n_estimators=300, max_features=0.3,
                                          random_state=0, n_jobs=1),
    }
    print(f"  {'model':<14} {'R^2':>9} {'MAE':>9} {'o_rt barglar':>14}")
    for nom, m in modellar.items():
        m.fit(Xtr, ytr)
        pred = m.predict(Xte)
        barglar = np.mean([e.get_n_leaves() for e in m.estimators_])
        print(f"  {nom:<14} {r2_score(yte, pred):>9.4f} "
              f"{mean_absolute_error(yte, pred):>9.4f} {barglar:>14.1f}")

    print("\n=== 2. max_features bo'yicha ikkalasi ===")
    print(f"  {'max_features':>13} {'RF R^2':>9} {'ET R^2':>9}")
    for mf in [1.0, 0.5, 0.3, "sqrt"]:
        r = cross_val_score(RandomForestRegressor(n_estimators=150,
                                                  max_features=mf,
                                                  random_state=0),
                            Xtr, ytr, cv=cv, scoring="r2").mean()
        e = cross_val_score(ExtraTreesRegressor(n_estimators=150,
                                                max_features=mf,
                                                random_state=0),
                            Xtr, ytr, cv=cv, scoring="r2").mean()
        print(f"  {str(mf):>13} {r:>9.4f} {e:>9.4f}")

    print("\n=== 3. Shovqin darajasi bo'yicha ===")
    print(f"  {'shovqin':>8} {'RF R^2':>9} {'ET R^2':>9} {'farq':>9}")
    for sh in [0.3, 1.0, 2.0, 4.0]:
        Xn, yn, _ = yarat(shovqin=sh, n=2000)
        r = cross_val_score(RandomForestRegressor(n_estimators=150,
                                                  max_features=0.3,
                                                  random_state=0),
                            Xn, yn, cv=cv, scoring="r2").mean()
        e = cross_val_score(ExtraTreesRegressor(n_estimators=150,
                                                max_features=0.3,
                                                random_state=0),
                            Xn, yn, cv=cv, scoring="r2").mean()
        print(f"  {sh:>8.1f} {r:>9.4f} {e:>9.4f} {e - r:>+9.4f}")

    print("\n=== 4. Model hajmi ===")
    for nom, m in modellar.items():
        tugunlar = sum(e.tree_.node_count for e in m.estimators_)
        print(f"  {nom:<14}: jami {tugunlar:,} tugun "
              f"(~{tugunlar * 48 / 1024 / 1024:.1f} MB)")
    print("  ⭐ Ansambl hajmi - ishlab chiqarishda muhim omil")


if __name__ == "__main__":
    main()

=== 1. Uch ansambl ===
  model                R^2       MAE   o_rt barglar
  bagging           0.6362    1.3277         1327.0
  RandomForest      0.6098    1.3536         1327.0
  ExtraTrees        0.5843    1.4011         2100.0

=== 2. max_features bo'yicha ikkalasi ===
   max_features    RF R^2    ET R^2
            1.0    0.6674    0.6817
            0.5    0.6595    0.6382
            0.3    0.6273    0.5839
           sqrt    0.5828    0.5223

=== 3. Shovqin darajasi bo'yicha ===
   shovqin    RF R^2    ET R^2      farq
       0.3    0.7530    0.7207   -0.0323
       1.0    0.6453    0.6200   -0.0252
       2.0    0.4370    0.4192   -0.0178
       4.0    0.1820    0.1730   -0.0089

=== 4. Model hajmi ===
  bagging       : jami 795,874 tugun (~36.4 MB)
  RandomForest  : jami 795,874 tugun (~36.4 MB)
  ExtraTrees    : jami 1,259,700 tugun (~57.7 MB)
  ⭐ Ansambl hajmi - ishlab chiqarishda muhim omil

ExtraTreesClassifier(oob_score=True)          # bootstrap=False  # ⚠️
ExtraTreesClassifier(bootstrap=True, oob_score=True)             # ✅

roc_auc_score(y, o.oob_decision_function_[:, 1])                 # ⚠️
m = ~np.isnan(oob); roc_auc_score(y[m], oob[m])                  # ✅

o.set_params(n_estimators=50).fit(X, y)   # 200 dan keyin        # ⚠️
o.set_params(n_estimators=400).fit(X, y)  # faqat oshirish       # ✅

model = RandomForestClassifier(n_estimators=500)                 # ⚠️
# RF va ET ni CV da taqqoslang                                   # ✅

RandomForestClassifier(n_estimators=500)   # 1 mln qator         # ⚠️
RandomForestClassifier(n_estimators=500, max_samples=0.3)        # ✅

o.oob_score_                                                     # ⚠️
cross_val_score(o, X, y, cv=TimeSeriesSplit(5))                  # ✅

RandomForestClassifier(n_estimators=2000)   # yuzlab MB          # ⚠️
# min_samples_leaf bilan hajmni cheklang                         # ✅

1.  # ET ning RF dan farqi?
2.  # ET da bootstrap standart qiymati?
3.  # ET nega tezroq?
4.  # ET qachon ustun?
5.  # rho ga ta'siri?
6.  # bias ga ta'siri?
7.  # ET da OOB uchun nima kerak?
8.  # warm_start nima qiladi?
9.  # oob_decision_function_ nima?
10. # NaN qachon paydo bo'ladi?
11. # max_samples nima uchun?
12. # RF da bootstrap=False?

1.  ExtraTreesClassifier(oob_score=True)

2.  roc_auc_score(y, o.oob_decision_function_[:, 1])

3.  o.set_params(n_estimators=50).fit(X, y)   # 200 dan keyin

4.  RandomForestClassifier(n_estimators=500)  # 2 mln qator

5.  o.oob_score_   # vaqt qatori