写一个代码按照小类别将数据集划分为训练集和测试集的方法，以确保训练集和测试集中每个小类别的比例相同

假设我们有一个数据集，每个样本都有一个类别标签。我们可以按照以下步骤划分训练集和测试集，并确保每个小类别的比例相同：

1. 对于每个小类别，计算样本数目，然后将其分为训练集和测试集。例如，我们可以将每个小类别的样本数的80％用于训练集，20％用于测试集。

2. 确保每个小类别的训练集和测试集的比例相同。如果某个小类别的样本数很少，可能需要将其与其他小类别合并。

下面是一个Python代码示例，它按照上述步骤将数据集划分为训练集和测试集：

import numpy as np

def split_dataset(X, y, test_ratio=0.2):
    """
    Split dataset into training and testing sets, ensuring that
    each small category has the same ratio in both sets.
    """
    # Determine the unique small categories
    small_categories = np.unique(y)

    # Split each small category into training and testing sets
    X_train, y_train, X_test, y_test = [], [], [], []
    for cat in small_categories:
        # Find samples belonging to this small category
        idx = np.where(y == cat)[0]
        n_samples = len(idx)

        # Split into training and testing sets
        n_test = int(n_samples * test_ratio)
        n_train = n_samples - n_test
        np.random.shuffle(idx)
        idx_train = idx[:n_train]
        idx_test = idx[n_train:n_samples]

        # Add to training and testing sets
        X_train.append(X[idx_train])
        y_train.append(y[idx_train])
        X_test.append(X[idx_test])
        y_test.append(y[idx_test])

    # Concatenate training and testing sets
    X_train = np.concatenate(X_train, axis=0)
    y_train = np.concatenate(y_train, axis=0)
    X_test = np.concatenate(X_test, axis=0)
    y_test = np.concatenate(y_test, axis=0)

    # Shuffle the training set
    idx = np.arange(len(y_train))
    np.random.shuffle(idx)
    X_train = X_train[idx]
    y_train = y_train[idx]

    return X_train, y_train, X_test, y_test

该函数接受输入数据集X和标签y，以及可选的测试集比例。它首先确定数据集中的小类别，然后将每个小类别分成训练集和测试集。最后，它将训练集和测试集串联起来，并对训练集进行随机排序。

使用示例：

X_train, y_train, X_test, y_test = split_dataset(X, y, test_ratio=0.2)