Cheat sheet va protokol kartasi

import pandas as pd

show(pd.DataFrame([
    {"#": 1, "Qadam": "Faylni och, shaklini ko'r", "Kod": "df.shape, df.head(), df.info()"},
    {"#": 2, "Qadam": "Target ni aniqla", "Kod": "df[target].value_counts()"},
    {"#": 3, "Qadam": "Vazifa turini belgila", "Kod": "sinf soni -> klassifikatsiya / regressiya"},
    {"#": 4, "Qadam": "Bo'sh qiymatlarni sana", "Kod": "df.isna().sum()"},
    {"#": 5, "Qadam": "Tuzoq ustunlarni tashla", "Kod": "ID, leakage, matn-son"},
    {"#": 6, "Qadam": "SPLIT (hammasidan oldin)", "Kod": "train_test_split(stratify=y)"},
    {"#": 7, "Qadam": "Baza qur", "Kod": "DummyClassifier / DummyRegressor"},
    {"#": 8, "Qadam": "Pipeline bilan birinchi model", "Kod": "Pipeline([tayyorlash, model])"},
    {"#": 9, "Qadam": "To'g'ri metrikani tanla", "Kod": "F1 / PR-AUC / MAE"},
    {"#": 10, "Qadam": "Cross-validation", "Kod": "cross_val_score(cv=5)"},
    {"#": 11, "Qadam": "Xato tahlili", "Kod": "qaysi qatorlarda xato qilyapti?"},
    {"#": 12, "Qadam": "Saqla va hujjatlashtir", "Kod": "joblib + manifest + DATASHEET"},
]))

import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.dummy import DummyClassifier

df = pd.read_csv("yangi.csv")
print(df.shape)
print(df.dtypes.to_string())
print(df.isna().sum().to_string())
print(df[TARGET].value_counts(normalize=True))

# har ustunda nechta noyob qiymat — ID va matn-son tuzoqlarini ochadi
for c in df.columns:
    print(f"{c:<24} noyob={df[c].nunique():<6} tur={df[c].dtype}")

y = df[TARGET]
X = df.drop(columns=[TARGET])
Xtr, Xte, ytr, yte = train_test_split(X, y, test_size=0.2,
                                      random_state=42, stratify=y)
baza = DummyClassifier(strategy="most_frequent").fit(Xtr, ytr)
print("BAZA:", baza.score(Xte, yte))

from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline
from sklearn.impute import SimpleImputer
from sklearn.preprocessing import StandardScaler, OneHotEncoder

SON = X.select_dtypes("number").columns.tolist()
MATN = X.select_dtypes("object").columns.tolist()

tayyorlash = ColumnTransformer([
    ("son", Pipeline([("imp", SimpleImputer(strategy="median")),
                      ("sc", StandardScaler())]), SON),
    ("matn", Pipeline([("imp", SimpleImputer(strategy="most_frequent")),
                       ("oh", OneHotEncoder(handle_unknown="ignore"))]), MATN),
])

quvur = Pipeline([("tayyorlash", tayyorlash), ("model", MODEL)])
quvur.fit(Xtr, ytr)

show(pd.DataFrame([
    {"Vaziyat": "Jadval, kam ma'lumot (<1000)", "Birinchi tanlov": "LogisticRegression / Ridge",
     "Keyin": "RandomForest"},
    {"Vaziyat": "Jadval, ko'p ma'lumot", "Birinchi tanlov": "LightGBM / XGBoost",
     "Keyin": "stacking"},
    {"Vaziyat": "Ko'p kategorik ustun", "Birinchi tanlov": "CatBoost",
     "Keyin": "target encoding + boosting"},
    {"Vaziyat": "Matn", "Birinchi tanlov": "TF-IDF + LogisticRegression",
     "Keyin": "fine-tuned transformer"},
    {"Vaziyat": "Rasm", "Birinchi tanlov": "tayyor CNN (transfer learning)",
     "Keyin": "fine-tuning"},
    {"Vaziyat": "Vaqt qatori", "Birinchi tanlov": "naive baza + lag + Ridge",
     "Keyin": "boosting (differensiyalangan)"},
    {"Vaziyat": "Tushuntirish majburiy", "Birinchi tanlov": "chiziqli model / bitta daraxt",
     "Keyin": "SHAP bilan boosting"},
]))

show(pd.DataFrame([
    {"Holat": "Klassifikatsiya, muvozanatli", "Metrika": "accuracy, ROC-AUC"},
    {"Holat": "Klassifikatsiya, nomutanosib", "Metrika": "F1, PR-AUC, recall@precision"},
    {"Holat": "Xato narxi assimetrik", "Metrika": "maxsus: FN va FP ga vazn"},
    {"Holat": "Regressiya, chetdagi qiymat bor", "Metrika": "MAE (RMSE emas)"},
    {"Holat": "Regressiya, katta xato juda yomon", "Metrika": "RMSE"},
    {"Holat": "Nisbiy xato muhim", "Metrika": "MAPE (nolga ehtiyot bo'ling)"},
    {"Holat": "Reyting / tartib", "Metrika": "NDCG, MAP"},
]))

show(pd.DataFrame([
    {"Belgi": "Train past, test past", "Tashxis": "underfitting",
     "Harakat": "murakkabroq model, ko'proq feature, kamroq regularizatsiya"},
    {"Belgi": "Train yuqori, test past", "Tashxis": "overfitting",
     "Harakat": "ko'proq ma'lumot, regularizatsiya, soddaroq model"},
    {"Belgi": "Ikkisi ham juda yuqori", "Tashxis": "LEAKAGE ehtimoli",
     "Harakat": "feature importance ga qara, ustunni olib tashlab o'lcha"},
    {"Belgi": "CV tarqoq (katta std)", "Tashxis": "ma'lumot kam yoki bir xil emas",
     "Harakat": "ko'proq burma, stratifikatsiya, ko'proq ma'lumot"},
    {"Belgi": "Bitta sinf umuman topilmaydi", "Tashxis": "nomutanosiblik",
     "Harakat": "class_weight, threshold, metrikani almashtir"},
    {"Belgi": "Offline yaxshi, prod yomon", "Tashxis": "drift yoki zanjir farqi",
     "Harakat": "kirish taqsimotini solishtir, Pipeline bir xilmi tekshir"},
]))

show(pd.DataFrame([
    {"Mavzu": "Split", "Eslatma": "Hamma narsadan OLDIN. stratify=y klassifikatsiyada"},
    {"Mavzu": "Scaler", "Eslatma": "fit faqat train da. Pipeline ichida bo'lsa — muammo yo'q"},
    {"Mavzu": "Vaqt qatori", "Eslatma": "Xronologik split, TimeSeriesSplit, differensiyalash"},
    {"Mavzu": "Guruhlangan ma'lumot", "Eslatma": "GroupShuffleSplit — bitta obyekt bitta bo'lakda"},
    {"Mavzu": "Daraxtlar", "Eslatma": "Scaling kerak emas"},
    {"Mavzu": "KNN, SVM, NN", "Eslatma": "Scaling SHART"},
    {"Mavzu": "random_state", "Eslatma": "Har joyda qo'ying — takrorlanish uchun"},
    {"Mavzu": "Neyron tarmoq", "Eslatma": "Kamida 3 seed, eng yomonini ham yozing"},
    {"Mavzu": "Test to'plami", "Eslatma": "Bir marta. Ikkinchi marta qarasangiz — u val"},
    {"Mavzu": "Saqlash", "Eslatma": "Pipeline + manifest + aylanma sinov"},
]))

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

from sklearn.model_selection import (train_test_split, cross_val_score,
                                     GridSearchCV, StratifiedKFold,
                                     GroupShuffleSplit, TimeSeriesSplit)
from sklearn.pipeline import Pipeline
from sklearn.compose import ColumnTransformer
from sklearn.impute import SimpleImputer
from sklearn.preprocessing import StandardScaler, OneHotEncoder
from sklearn.dummy import DummyClassifier, DummyRegressor
from sklearn.linear_model import LogisticRegression, Ridge, Lasso
from sklearn.ensemble import RandomForestClassifier, GradientBoostingClassifier
from sklearn.metrics import (accuracy_score, precision_score, recall_score,
                             f1_score, roc_auc_score, average_precision_score,
                             confusion_matrix, classification_report,
                             mean_absolute_error, r2_score)
import joblib

show(pd.DataFrame([
    {"Qoida": "1. Baza birinchi", "Nega": "Baza bo'lmasa, natija ma'nosiz"},
    {"Qoida": "2. Split hammasidan oldin", "Nega": "Keyin qilingan har amal leakage xavfi"},
    {"Qoida": "3. Pipeline majburiy", "Nega": "Leakage ni tuzilish darajasida to'xtatadi"},
    {"Qoida": "4. Har raqam o'lchangan bo'lsin", "Nega": "Taxmin qilingan raqam — yolg'on"},
    {"Qoida": "5. Taxminingiz rad etilsa, yozib qo'ying", "Nega": "Eng qimmatli bilim shu yerda"},
]))
print()
print("Bu kitobda men bir necha marta xato taxmin qildim va o'lchov meni rad etdi:")
print("CNN markazlashgan rasmda ustun kelmadi, LSTM beqaror chiqdi, TF-IDF")
print("transformerdan yuqori bo'ldi, ustun tartibi muhim emas ekan.")
print()
print("Har safar men taxminni emas, O'LCHOVNI yozdim. Siz ham shunday qiling.")