机器学习基础

# 图灵测试：如果你隔着屏幕分不清对方是人还是机器，
# 那么这台机器就「会思考」。
qa = [
    ("你几岁？", "我今年 12 岁零 3 个版本号～"),
    ("1+1=?", "2。要不要听个冷笑话？"),
    ("你会写诗吗", "会！押韵的那种。"),
]
for q, a in qa:
    print(f"问：{q}\n答：{a}")
print("AI 三次浪潮：符号主义 → 机器学习 → 深度学习/大模型")

from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
from sklearn.tree import DecisionTreeClassifier

X, y = load_iris(return_X_y=True)       # 150 朵花 × 4 个特征
Xtr, Xte, ytr, yte = train_test_split(X, y, test_size=0.3, random_state=1)
model = DecisionTreeClassifier().fit(Xtr, ytr)
print("测试集准确率：%.2f" % model.score(Xte, yte))
print("有标签→监督学习；无标签聚类→无监督学习")

from sklearn.linear_model import LinearRegression
# 面积(㎡) → 房价(万)
X = [[50], [80], [100], [120]]
y = [150, 230, 290, 350]
model = LinearRegression().fit(X, y)
print("每平米约：", round(model.coef_[0], 1), "万")
print("预测 90㎡：", round(model.predict([[90]])[0], 1), "万")

import math
def sigmoid(z):
    return 1 / (1 + math.exp(-z))

for z in [-2, 0, 2]:
    print(f"z={z:+d} → 概率 {sigmoid(z):.2f}")
print("≥0.5 判为垃圾邮件，<0.5 判为正常邮件")
# 决策边界：z = w·x + b = 0 的那条线

from sklearn.tree import DecisionTreeClassifier, export_text
# 特征: [有无羽毛, 会飞吗]  标签: 0鸟 1猫 2鱼
X = [[1,1],[1,0],[0,0],[0,0],[0,1]]
y = [0, 0, 1, 2, 2]
tree = DecisionTreeClassifier(max_depth=2).fit(X, y)
print(export_text(tree,
      feature_names=["羽毛", "会飞"]))

# 随机森林 = 多棵「各有偏见」的决策树 + 投票
from sklearn.ensemble import RandomForestClassifier
from sklearn.datasets import load_digits
X, y = load_digits(return_X_y=True)
model = RandomForestClassifier(n_estimators=100, random_state=1)
model.fit(X[:1400], y[:1400])
print("手写数字准确率：%.3f" % model.score(X[1400:], y[1400:]))
print("两重随机：①有放回抽样本 ②每次分裂只看部分特征")
print("防止所有树长成一个样 → 投票才有意义")

# KNN：看身边 K 个邻居，少数服从多数
train = [(1, "动作片"), (2, "动作片"),
         (8, "爱情片"), (9, "爱情片")]
new = 3
k = 3
dists = sorted(train, key=lambda t: abs(t[0] - new))[:k]
votes = {}
for _, label in dists:
    votes[label] = votes.get(label, 0) + 1
print("新样本 =", new, "→ 预测：", max(votes, key=votes.get))
print("k 太小怕噪声，k 太大怕混淆，要试出来的")

import numpy as np
from sklearn.cluster import KMeans
# 假设一张图的 6 个像素（RGB）
pixels = np.array([[250,10,10],[240,20,20],[10,10,250],
                   [20,20,240],[120,120,120],[130,130,130]])
kmeans = KMeans(n_clusters=3).fit(pixels)   # 收敛到 3 种主色
for c in kmeans.cluster_centers_.astype(int):
    print("主色 RGB：", c)

from sklearn.metrics import (
    accuracy_score, precision_score, recall_score)
y_true = [1, 0, 1, 1, 0, 1]     # 真实答案
y_pred = [1, 0, 0, 1, 0, 1]     # 模型预测
print("准确率：", accuracy_score(y_true, y_pred))
print("精确率：", precision_score(y_true, y_pred))
print("召回率：", recall_score(y_true, y_pred))

import numpy as np
# 三大件：造特征 / 挑特征 / 变尺度
x = np.array([[180, 70], [165, 55], [175, 80]])
# ① 造特征：身高体重 → BMI（更有信息量）
bmi = x[:, 1] / (x[:, 0] / 100) ** 2
# ② 标准化：均值 0、方差 1，防止量纲偏心
z = (x - x.mean(0)) / x.std(0)
print("BMI：", np.round(bmi, 1))
print("标准化后第1行：", np.round(z[0], 2))

from sklearn.pipeline import make_pipeline
from sklearn.impute import SimpleImputer
from sklearn.ensemble import GradientBoostingClassifier

# ① 读数据 → ② 特征工程 → ③ 划分 → ④ 建管道 → ⑤ 训练 → ⑥ 评估
pipe = make_pipeline(
    SimpleImputer(strategy="median"),      # 自动填缺失值
    GradientBoostingClassifier(random_state=1))
# pipe.fit(X_train, y_train)
# print(pipe.score(X_test, y_test))       # 典型 ~0.80
print("管道 = 预处理 + 模型 一体化，防数据泄漏")
print("Kaggle 泰坦尼克：经典入门赛，新手第一站")