快速入门指南

恭喜您成功安装了Sequenzo！ 🎉 现在就可以开始玩转社会序列分析啦。

这个指南会手把手教你分析国家级CO₂排放序列。想了解我们是怎么把原始数据处理成适合做序列分析的格式？详细过程看这里（TODO：SequenzoWebsite网页）。

python新手？没关系！Sequenzo专门为小白设计，界面直观好上手，不管你是编程萌新还是Python老司机，都能快速掌握。

这个教程结束后，你将掌握：

安装Sequenzo
导入和对数据进行初步探索
分析社会序列
结果可视化

好了，开始我们的数据分析之旅吧！🐍✨

1. 初步了解数据

python

# 导入必要的库
# 你的调用代码（比如在脚本或notebook中）

from sequenzo import * # 导入包，给它一个简短的别名
import pandas as pd # 数据处理

# 列出Sequenzo中所有可用的数据集
# 现在用别名来调用函数：
print('Sequenzo中的可用数据集: ', list_datasets())

# 加载我们在本教程中要探索的数据
# `df`是`dataframe`的缩写，这是数据集的常用变量名
# df = load_dataset('country_co2_emissions')
df = load_dataset('country_co2_emissions_global_deciles')

# 显示数据框
df

Sequenzo中的可用数据集:  ['biofam', 'biofam_child_domain', 'biofam_left_domain', 'biofam_married_domain', 'chinese_colonial_territories', 'country_co2_emissions', 'country_co2_emissions_global_deciles', 'country_co2_emissions_global_quintiles', 'country_co2_emissions_local_deciles', 'country_co2_emissions_local_quintiles', 'country_gdp_per_capita', 'polyadic_samplec1', 'polyadic_samplep1', 'polyadic_seqc1', 'polyadic_seqp1']

	country	1800	1801	1802	1803	1804	1805	1806	1807	1808	...	2013	2014	2015	2016	2017	2018	2019	2020	2021	2022
0	Afghanistan	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	...	D7	D7	D7	D7	D7	D7	D7	D7	D7	D7
1	Albania	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	...	D9	D9	D9	D9	D9	D9	D9	D9	D9	D9
2	Algeria	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	...	D9	D9	D9	D9	D9	D9	D9	D9	D9	D9
3	Andorra	D7	D7	D7	D7	D7	D7	D7	D7	D7	...	D10 (Very High)	D10 (Very High)	D10 (Very High)	D10 (Very High)	D10 (Very High)	D10 (Very High)	D10 (Very High)	D10 (Very High)	D10 (Very High)	D10 (Very High)
4	Angola	D3	D3	D3	D3	D3	D3	D3	D3	D3	...	D8	D8	D8	D8	D8	D8	D8	D8	D8	D8
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
189	Venezuela	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	...	D9	D9	D9	D9	D9	D8	D6	D8	D8	D8
190	Vietnam	D3	D3	D3	D3	D3	D3	D3	D3	D3	...	D8	D8	D8	D8	D8	D9	D9	D9	D9	D9
191	Yemen	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	D1 (Very Low)	...	D8	D8	D7	D7	D7	D7	D7	D7	D7	D7
192	Zambia	D7	D7	D7	D7	D7	D7	D7	D7	D7	...	D7	D7	D7	D7	D7	D7	D7	D7	D7	D7
193	Zimbabwe	D5	D5	D5	D5	D5	D5	D5	D5	D5	...	D8	D8	D8	D8	D8	D8	D8	D8	D8	D8

194 rows × 224 columns

这个分类基于所有年份的人均CO₂排放值：

极低（后20%）
低（20-40%）
中等（40-60%）
高（60-80%）
极高（前20%）

用社会序列分析的术语来说，每个类别叫做一个状态，状态的序列叫做社会序列。总的来说，这个数据集展示了每个国家每年的人均CO₂排放水平。

以安道尔为例：

python

# 筛选安道尔的数据
andorra_df = df[df['country'] == 'Andorra']

# 显示安道尔的数据框
andorra_df

	country	1800	1801	1802	1803	1804	1805	1806	1807	1808	...	2013	2014	2015	2016	2017	2018	2019	2020	2021	2022
3	Andorra	D7	D7	D7	D7	D7	D7	D7	D7	D7	...	D10 (Very High)	D10 (Very High)	D10 (Very High)	D10 (Very High)	D10 (Very High)	D10 (Very High)	D10 (Very High)	D10 (Very High)	D10 (Very High)	D10 (Very High)

1 rows × 224 columns

根据结果（原数据集的一个子集），我们可以看到安道尔的人均CO₂排放在这些年里经历了以下变化：

1990-1991: 从"极高"水平开始，说明排放量在所有国家中排前20%
1992-1997: 下降到"高"水平（60-80百分位）
1998: 短暂回到"极高"水平
2000年代至今: 稳定在"高"水平（60-80百分位），并一直保持到2019年

但这就带来了一个问题：如果我们想分析所有国家的序列怎么办？ 🤔

这时候Sequenzo就能派上用场！

2. 使用 Sequenzo 分析社会序列

python

# 创建 SequenceData 对象

# 定义时间跨度变量
time_list = list(df.columns)[1:]

# states = ['Very Low', 'Low', 'Middle', 'High', 'Very High']
states = ['D1 (Very Low)', 'D10 (Very High)', 'D2', 'D3', 'D4', 'D5', 'D6', 'D7', 'D8', 'D9']

# TODO: 编写异常处理逻辑：如果参数不存在，则提示传入正确的参数
# sequence_data = SequenceData(df, time=time, id_col="country", ids=df['country'].values, states=states)

sequence_data = SequenceData(df,
                             time=time_list,
                             id_col="country",
                             states=states,
                             labels=states)

sequence_data

[!] Detected missing values (empty cells) in the sequence data.
    → Automatically added 'Missing' to `states` and `labels` for compatibility.
    However, it's strongly recommended to manually include it when defining `states` and `labels`.
    For example:

        states = ['At Home', 'Left Home', 'Missing']
        labels = ['At Home', 'Left Home', 'Missing']

    This ensures consistent color mapping and avoids unexpected visualization errors.

[>] SequenceData initialized successfully! Here's a summary:
[>] Number of sequences: 194
[>] Number of time points: 223
[>] Min/Max sequence length: 216 / 223
[>] There are 7 missing values across 1 sequences.
    First few missing sequence IDs: ['Panama'] ...
[>] Top sequences with the most missing time points:
    (Each row shows a sequence ID and its number of missing values)

             Missing Count
Sequence ID               
Panama                   7
[>] States: ['D1 (Very Low)', 'D10 (Very High)', 'D2', 'D3', 'D4', 'D5', 'D6', 'D7', 'D8', 'D9', 'Missing']
[>] Labels: ['D1 (Very Low)', 'D10 (Very High)', 'D2', 'D3', 'D4', 'D5', 'D6', 'D7', 'D8', 'D9', 'Missing']





SequenceData(194 sequences, States: ['D1 (Very Low)', 'D10 (Very High)', 'D2', 'D3', 'D4', 'D5', 'D6', 'D7', 'D8', 'D9', 'Missing'])

数据可视化

在众多可视化方法中，索引图是最常用的。下面我们来看看它的效果。

我们刚才已经成功创建了SequenceData对象，这是Sequenzo分析社会序列的核心工具。

如果只用肉眼看原始数据，我们一次只能关注一个国家的发展轨迹。但有了Sequenzo，我们就能同时分析所有国家的数据！可视化是其中最关键的功能，它能帮我们发现数据中隐藏的规律和趋势。

python

# 绘制索引图
# TODO: 在这里同样处理意外参数的问题。TypeError: plot_sequence_index() 函数收到了意外的关键字参数 'sortv'

plot_sequence_index(sequence_data,
                    # save_as='index_plot'
                    )

png

<Figure size 640x480 with 0 Axes>

还想挖掘更多信息？状态分布图来啦！

python

sequence_data.plot_legend(save_as="legend_plot")

png

python

plot_most_frequent_sequences(sequence_data, save_as='test', top_n=5)hh

png

<Figure size 640x480 with 0 Axes>

python

plot_mean_time(sequence_data, save_as='mean_time')

png

<Figure size 640x480 with 0 Axes>

python

plot_transition_matrix(sequence_data, save_as='transition_matrix')

png

<Figure size 640x480 with 0 Axes>

python

plot_state_distribution(sequence_data, save_as='state_distribution')

png

<Figure size 640x480 with 0 Axes>

python

plot_modal_state(sequence_data, save_as='modal_state')

png

差异性度量

python

# 参数替换选项：用"OM/DHD/HAM"代替"OMspell"，用"CONSTANT"代替"TRATE"
om = get_distance_matrix(seqdata=sequence_data,
                         method='OM',
                         sm="TRATE",
                         indel="auto")
om

[>] Processing 194 sequences with 11 unique states.
[>] Transition-based substitution-cost matrix (TRATE) initiated...
  - Computing transition probabilities for: [D1 (Very Low), D10 (Very High), D2, D3, D4, D5, D6, D7, D8, D9, Missing]
[>] Indel cost generated.

[>] Identified 194 unique sequences.
[>] Starting Optimal Matching(OM)...
[>] Computing all pairwise distances...
[>] Computed Successfully.

	Afghanistan	Albania	Algeria	Andorra	Angola	Antigua and Barbuda	Argentina	Armenia	Australia	Austria	...	Uganda	Ukraine	Uruguay	Uzbekistan	Vanuatu	Venezuela	Vietnam	Yemen	Zambia	Zimbabwe
Afghanistan	0.000000	272.363879	141.530885	405.839693	357.116339	352.452702	369.202821	199.747566	363.605396	367.308098	...	356.188303	251.681276	391.564474	253.570661	333.417093	193.512323	339.495876	337.004372	405.784131	389.419165
Albania	272.363879	0.000000	196.640891	293.042386	360.848766	357.623372	291.258820	190.293293	330.065835	318.787795	...	301.974406	207.816080	269.205424	205.916577	385.871215	217.202167	355.905637	150.343476	331.733788	324.072412
Algeria	141.530885	196.640891	0.000000	299.876169	360.346243	329.826237	297.445339	131.698913	285.521108	291.201892	...	386.745755	173.060640	276.896587	145.923477	361.337985	75.716405	348.517376	309.997189	397.970384	368.047516
Andorra	405.839693	293.042386	299.876169	0.000000	356.053656	317.797407	119.351333	288.455080	161.034886	143.550606	...	442.272407	231.894801	323.007193	256.673631	337.870059	320.299300	305.637267	338.668802	248.972441	316.165089
Angola	357.116339	360.848766	360.346243	356.053656	0.000000	291.024809	338.252215	304.404207	333.524614	343.035738	...	299.756320	323.190476	213.309453	329.042700	223.247060	388.304116	126.257652	324.332208	355.987397	347.278292
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
Venezuela	193.512323	217.202167	75.716405	320.299300	388.304116	332.208197	314.154419	167.770756	307.697612	294.576219	...	407.898441	178.451252	299.052292	159.381580	393.463182	0.000000	386.746863	338.328445	409.781349	363.126942
Vietnam	339.495876	355.905637	348.517376	305.637267	126.257652	277.169745	281.862722	262.901696	284.480681	296.840967	...	250.273278	313.199712	224.199999	317.111338	171.558720	386.746863	0.000000	332.362799	311.052135	313.268987
Yemen	337.004372	150.343476	309.997189	338.668802	324.332208	402.822666	338.626193	287.908848	368.726059	362.779293	...	210.976531	287.655467	289.098219	284.491015	338.798635	338.328445	332.362799	0.000000	323.373702	353.783341
Zambia	405.784131	331.733788	397.970384	248.972441	355.987397	400.131845	247.433419	362.309280	364.953012	359.363387	...	442.218057	373.720811	406.480220	371.778073	255.268940	409.781349	311.052135	323.373702	0.000000	304.666582
Zimbabwe	389.419165	324.072412	368.047516	316.165089	347.278292	279.071995	315.245317	292.686113	257.977877	282.863190	...	372.870754	325.264179	340.907764	320.059748	347.484712	363.126942	313.268987	353.783341	304.666582	0.000000

194 rows × 194 columns

python

plot_relative_frequency(seqdata=sequence_data,
                        distance_matrix=om,
                        num_groups=12,
                        dpi=200,
                        # save_as='relative_frequency_plot'
                        )

# 可视化主要问题
# 1. 当把横坐标从“严格的分类标签”变成“数值（年份/年龄）”时，刻度有可能跑偏，得手动把 xticks／布局调回来。
# 2. 状态（分类）多了以后，默认 legend 效果会很糟，得自己选位置或分列、扩增画布空间，才能让图例和主体同时都好看、好读。

png

<Figure size 640x480 with 0 Axes>

聚类分析

python

cluster = Cluster(om, sequence_data.ids, clustering_method='ward')
cluster.plot_dendrogram(xlabel="Countries", ylabel="Distance")

[>] Converting DataFrame to NumPy array...

png

<Figure size 640x480 with 0 Axes>

python

# Create a ClusterQuality object to evaluate clustering quality创建一个 ClusterQuality 对象，用于评估聚类质量。
cluster_quality = ClusterQuality(cluster)
cluster_quality.compute_cluster_quality_scores()
cluster_quality.plot_combined_scores(norm='zscore', save_as='combined_scores')
summary_table = cluster_quality.get_metrics_table()
print(summary_table)

png

  Metric  Opt. Clusters  Opt. Value  Z-Score Norm.  Min-Max Norm.
0    ASW              6    1.922696       1.922696            1.0
1   ASWw              6    1.922696       1.922696            1.0
2     HG              2    3.235991       3.235991            1.0
3    PBC             20    1.028963       1.028963            1.0
4     CH              8    0.613044       0.613044            1.0
5     R2             20    1.297366       1.297366            1.0
6     HC             20    1.228937       1.228937            1.0



<Figure size 640x480 with 0 Axes>

python

cluster_results = ClusterResults(cluster)
membership_table = cluster_results.get_cluster_memberships(num_clusters=5)
print(membership_table)
distribution = cluster_results.get_cluster_distribution(num_clusters=5)
print(distribution)
cluster_results.plot_cluster_distribution(num_clusters=5, save_as="distribution.png", title=None)

       Entity ID  Cluster
0    Afghanistan        1
1        Albania        1
2        Algeria        1
3        Andorra        3
4         Angola        4
..           ...      ...
189    Venezuela        1
190      Vietnam        4
191        Yemen        2
192       Zambia        3
193     Zimbabwe        5

[194 rows x 2 columns]
   Cluster  Count  Percentage
0        1     55       28.35
1        2     30       15.46
2        3     49       25.26
3        4     18        9.28
4        5     42       21.65


/Users/lei/Documents/Sequenzo_all_folders/Sequenzo-main/sequenzo/clustering/hierarchical_clustering.py:598: FutureWarning: 

Passing `palette` without assigning `hue` is deprecated and will be removed in v0.14.0. Assign the `x` variable to `hue` and set `legend=False` for the same effect.

  ax = sns.barplot(x='Cluster', y='Count', data=distribution, palette='pastel')

png

<Figure size 640x480 with 0 Axes>

python

mapping_dict = {
    1: 'xxx',
    2: 'Test',
    3: 'The name of the cluster',
    4: 'xxxxx',
    5: 'whsiaa',
}

membership_table = replace_cluster_id_by_labels(membership_table,
                                                mapping=mapping_dict,
                                                new_cluster_column_name='Cluster',
                                                new_id_column_name='ID')

membership_table

	ID	Cluster
0	Afghanistan	xxx
1	Albania	xxx
2	Algeria	xxx
3	Andorra	The name of the cluster
4	Angola	xxxxx
...	...	...
189	Venezuela	xxx
190	Vietnam	xxxxx
191	Yemen	Test
192	Zambia	The name of the cluster
193	Zimbabwe	whsiaa

194 rows × 2 columns

python

plot_sequence_index(seqdata=sequence_data,
                    id_group_df=membership_table,
                    categories='Cluster',
                    # save_as='cluster_index_plot'
                    )

png

python

plot_state_distribution(seqdata=sequence_data,
                            id_group_df=membership_table,
                            categories='Cluster',
                            # save_as='cluster_state_distribution_plot'
                        )

png

Regression

python

Conclusions

python

快速入门指南 ​

1. 初步了解数据 ​

2. 使用 Sequenzo 分析社会序列 ​

数据可视化 ​

差异性度量 ​

聚类分析 ​

Regression ​

Conclusions ​

快速入门指南

1. 初步了解数据

2. 使用 Sequenzo 分析社会序列

数据可视化

差异性度量

聚类分析

Regression

Conclusions