图片详情
—
Skip to content

check_esol_comparisons.py ​

返回章节 · 配套代码说明

复制以下内容并保存为 check_esol_comparisons.py,使用方法见配套代码说明。

python
"""复核 ESOL 案例的多随机种子结果与充分优化的线性基线。

在本目录运行:python3 check_esol_comparisons.py
依赖 NumPy 求最小二乘解;网络训练使用同目录的纯 Python MLP。
"""

from pathlib import Path
from statistics import mean, stdev

import numpy as np

from demo_delaney_mlp import (
    SimpleStandardScaler,
    load_delaney_data,
    r2_score,
    rmse_score,
    split_train_test,
)
from nn import MLP


def main():
    rows = load_delaney_data(Path(__file__).with_name("delaney-processed.csv"))
    fields = [
        "Molecular Weight",
        "Number of H-Bond Donors",
        "Number of Rings",
        "Number of Rotatable Bonds",
        "Polar Surface Area",
    ]
    x = [[float(row[field]) for field in fields] for row in rows]
    y = [[float(row["measured log solubility in mols per litre"])] for row in rows]
    x_train, x_test, y_train, y_test = split_train_test(x, y, test_ratio=0.2, seed=42)
    scaler = SimpleStandardScaler()
    x_train = scaler.fit_transform(x_train)
    x_test = scaler.transform(x_test)
    targets = [value[0] for value in y_test]

    r2_values, rmse_values = [], []
    for seed in range(5):
        net = MLP([5, 16, 8, 1], hidden_activation="tanh",
                  output_activation="linear", loss="mse", seed=seed)
        net.train(x_train, y_train, epochs=300, lr=0.02,
                  batch_size=32, momentum=0.9, seed=seed)
        predictions = [net.predict(row)[0] for row in x_test]
        r2_values.append(r2_score(targets, predictions))
        rmse_values.append(rmse_score(targets, predictions))
        print(f"seed={seed}: R²={r2_values[-1]:.3f}, RMSE={rmse_values[-1]:.3f}",
              flush=True)
    print(f"R² 均值 ± 样本标准差:{mean(r2_values):.3f} ± {stdev(r2_values):.3f}")
    print(f"RMSE 均值 ± 样本标准差:{mean(rmse_values):.3f} ± {stdev(rmse_values):.3f}")

    train_design = np.column_stack([x_train, np.ones(len(x_train))])
    test_design = np.column_stack([x_test, np.ones(len(x_test))])
    coefficients, *_ = np.linalg.lstsq(
        train_design, np.asarray(y_train).ravel(), rcond=None
    )
    predictions = [
        sum(float(value) * float(weight) for value, weight in zip(row, coefficients))
        for row in test_design
    ]
    print(f"最小二乘线性基线:R²={r2_score(targets, predictions):.3f}, "
          f"RMSE={rmse_score(targets, predictions):.3f}")


if __name__ == "__main__":
    main()

《原子智能》· 纸质出版预备版 · PolyAI Team 著