计算Pandas数据框间Cohen's Kappa一致性系数教程

import pandas as pd
import numpy as np
from sklearn.metrics import cohen_kappa_score

# 示例数据框1
data1 = {'subject': ['A', 'B', 'C', 'D'],
         'group': ['red', 'red', 'blue', 'blue'],
         'lists': [[0, 1, 1], [0, 0, 0], [1, 1, 1], [0, 1, 0]]}
df1 = pd.DataFrame(data1)

# 示例数据框2
data2 = {'subject': ['a', 'b', 'c', 'd'],
         'group': ['red', 'red', 'blue', 'blue'],
         'lists': [[0, 1, 0], [1, 1, 0], [1, 0, 1], [1, 1, 0]]}
df2 = pd.DataFrame(data2)

print("DataFrame 1:")
print(df1)
print("\nDataFrame 2:")
print(df2)

# 计算 df1 中所有列表与 df2 中所有列表的 pairwise kappa
kappa_scores_flat = [cohen_kappa_score(i, j) for i in df1['lists'] for j in df2['lists']]
print("\nPairwise Kappa Scores (df1 vs df2, flat list):")
print(kappa_scores_flat)

# 使用 itertuples() 构建 df1 (行) vs df2 (列) 的 Kappa 矩阵
kappa_matrix_df1_vs_df2 = pd.DataFrame(
    {df2_row.subject: {df1_row.subject: cohen_kappa_score(df1_row.lists, df2_row.lists)
                       for df1_row in df1.itertuples()}
     for df2_row in df2.itertuples()}
).T # 转置以使 df1 subjects 为行，df2 subjects 为列

print("\nPairwise Kappa Matrix (df1 subjects as rows, df2 subjects as columns):")
print(kappa_matrix_df1_vs_df2)

# 创建 df1 和 df2 的笛卡尔积
cross_merged_df = df1.merge(df2, how='cross', suffixes=('_df1', '_df2'))

# 提取 'lists' 列，并使用 np.vectorize 计算 kappa
# 注意：cohen_kappa_score 期望一维数组，所以需要先转换为 numpy 数组再转置
kappa_scores_vectorized = np.vectorize(cohen_kappa_score)(
    *cross_merged_df[['lists_df1', 'lists_df2']].to_numpy().T
)

# 将结果添加到交叉合并的 DataFrame 中
cross_merged_df['kappa_score'] = kappa_scores_vectorized

print("\nCross-merged DataFrame with Kappa Scores:")
print(cross_merged_df[['subject_df1', 'subject_df2', 'kappa_score']])

# 1. 合并所有受试者及其数据到一个数据框
# 添加一个 'source' 列以区分来源，如果需要的话
df_combined = pd.concat([
    df1.assign(source='df1'),
    df2.assign(source='df2')
]).reset_index(drop=True)

# 2. 构建完整的 pairwise Kappa 矩阵
# 行和列都包含 df_combined 中的所有受试者
kappa_matrix_full = pd.DataFrame(
    {row_i.subject: {row_j.subject: cohen_kappa_score(row_i.lists, row_j.lists)
                     for row_j in df_combined.itertuples()}
     for row_i in df_combined.itertuples()}
)

print("\nComprehensive Pairwise Kappa Matrix (all subjects):")
print(kappa_matrix_full)

import seaborn as sns
import matplotlib.pyplot as plt

# 为了更好的可视化，可以根据 'group' 列对矩阵进行排序
# 首先，创建一个包含所有subject及其group的Series
subject_groups = pd.concat([df1.set_index('subject')['group'], df2.set_index('subject')['group']])
sorted_subjects = subject_groups.sort_values().index.tolist()

# 重新排序 kappa_matrix_full
kappa_matrix_sorted = kappa_matrix_full.loc[sorted_subjects, sorted_subjects]

plt.figure(figsize=(10, 8))
sns.heatmap(kappa_matrix_sorted, annot=True, cmap='viridis', fmt=".2f", linewidths=.5)
plt.title("Pairwise Cohen's Kappa Agreement Heatmap (Sorted by Group)")
plt.xlabel("Subject 2")
plt.ylabel("Subject 1")
plt.show()