机器学习项目闭环：从问题定义到可复现预测

使用者：谁根据预测采取行动？
预测对象：一行样本代表什么？
预测时点：什么时候产生预测？
输入：该时点已经知道哪些字段？
目标：预测什么结果、观察窗口多长？
指标：哪类错误代价更高？
基线：当前规则或最简单策略是什么？

from dataclasses import asdict, dataclass


@dataclass
class Experiment:
    data_version: str
    random_state: int
    model: str
    validation_f1: float


record = Experiment("batch-v1", 42, "logistic-regression", 0.71)
print(asdict(record))

python -m pip install scikit-learn

from sklearn.datasets import make_classification
from sklearn.dummy import DummyClassifier
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import f1_score
from sklearn.model_selection import train_test_split

X, y = make_classification(
    n_samples=800, n_features=8, n_informative=5,
    weights=[0.75, 0.25], random_state=42,
)
X_train, X_temp, y_train, y_temp = train_test_split(
    X, y, test_size=0.4, stratify=y, random_state=42,
)
X_valid, X_test, y_valid, y_test = train_test_split(
    X_temp, y_temp, test_size=0.5, stratify=y_temp, random_state=42,
)

baseline = DummyClassifier(strategy="most_frequent").fit(X_train, y_train)
model = LogisticRegression(max_iter=1000, random_state=42).fit(X_train, y_train)
print("baseline valid F1:", f1_score(y_valid, baseline.predict(X_valid)))
print("model valid F1:", round(f1_score(y_valid, model.predict(X_valid)), 3))
print("final test F1:", round(f1_score(y_test, model.predict(X_test)), 3))

from sklearn.model_selection import learning_curve
import numpy as np

sizes, train_scores, valid_scores = learning_curve(
    LogisticRegression(max_iter=1000),
    X_train,
    y_train,
    cv=5,
    scoring="f1",
    train_sizes=np.linspace(0.2, 1.0, 5),
)
for size, train_mean, valid_mean in zip(
    sizes, train_scores.mean(axis=1), valid_scores.mean(axis=1)
):
    print(size, round(train_mean, 3), round(valid_mean, 3))