如何用pandas.concat复刻已废弃append方法的原有行为？

我接手的代码使用了pandas的append方法，触发了官方弃用警告：

The frame.append method is deprecated and will be removed from pandas
in a future version. Use pandas.concat instead.

尝试改用pandas.concat保留原有append行为，但多次尝试均失败：创建一个(0,31)的空DataFrame，用append添加一条空行后结果为(1,31)，但各种concat写法都得到(1,32)的结果。

复现代码 #

import pandas as pd

# 创建带列名的空DataFrame
obs = pd.DataFrame(columns=['basedatetime_before', 'lat_before', 'lon_before', 
                            'sog_before', 
                            'cog_before', 
                            'heading_before', 
                            'vesselname_before', 'imo_before', 
                            'callsign_before', 
                            'vesseltype_before', 'status_before', 
                            'length_before', 'width_before', 
                            'draft_before',
                            'cargo_before', 
                            'basedatetime_after', 'lat_after', 
                            'lon_after', 
                            'sog_after', 
                            'cog_after', 'heading_after', 
                            'vesselname_after', 'imo_after', 
                            'callsign_after', 
                            'vesseltype_after', 'status_after', 
                            'length_after', 'width_after', 
                            'draft_after', 
                            'cargo_after'])

# 初始化DataFrame
desired = pd.Timestamp('2016-03-20 00:05:00+0000', tz='UTC')
obs['point'] = desired
obs['basedatetime_before'] = pd.to_datetime(obs['basedatetime_before'])
obs['basedatetime_after'] = pd.to_datetime(obs['basedatetime_after'])
obs.rename(lambda s: s.lower(), axis = 1, inplace = True)

# 创建新的"空行"Series
new_obs = pd.Series([desired], index=['point'])

# 打印初始形状
print("Orig obs.shape", obs.shape)
print("New_obs.shape", new_obs.shape)
print("--------------------------------------")

# 原append写法（正常得到(1,31)）
obs1 = obs.append(new_obs, ignore_index=True)

# 各种尝试的concat写法（均得到(1,32)）
obs2 = pd.concat([obs, new_obs])
obs3 = pd.concat([obs, new_obs], ignore_index=True)
obs4 = pd.concat([obs, new_obs.T])
obs5 = pd.concat([obs, new_obs.T], ignore_index=True)
obs6 = pd.concat([new_obs, obs])
obs7 = pd.concat([new_obs, obs], ignore_index=True)
obs8 = pd.concat([new_obs.T, obs])
obs9 = pd.concat([new_obs.T, obs], ignore_index=True)

# 验证append仍正常工作
obs10 = obs.append(new_obs, ignore_index=True)

# 打印结果
print("----> obs1.shape",obs1.shape)
print("obs2.shape",obs2.shape)
print("obs3.shape",obs3.shape)
print("obs4.shape",obs4.shape)
print("obs5.shape",obs5.shape)
print("obs6.shape",obs6.shape)
print("obs7.shape",obs7.shape)
print("obs8.shape",obs8.shape)
print("obs9.shape",obs9.shape)
print("----> obs10.shape",obs10.shape)

运行结果 #

Orig obs.shape (0, 31)
New_obs.shape (1,)
--------------------------------------
----> obs1.shape (1, 31)
obs2.shape (1, 32)
obs3.shape (1, 32)
obs4.shape (1, 32)
obs5.shape (1, 32)
obs6.shape (1, 32)
obs7.shape (1, 32)
obs8.shape (1, 32)
obs9.shape (1, 32)
----> obs10.shape (1, 31) 

问题根源：直接拼接Series和DataFrame时，concat会将Series的索引当作新列，导致列数增加。而append会自动将Series视为一行，按原DataFrame的列对齐，缺失列填充NaN。

要实现和append完全一致的效果，需先将Series转换为与原DataFrame列对齐的单行DataFrame，再进行拼接：

方法1：先对齐列再转换为DataFrame #

# 将new_obs按原DataFrame的列重新索引，缺失列自动填充NaN
aligned_new_obs = new_obs.reindex(obs.columns)
# 转换为单行DataFrame后拼接，设置ignore_index=True（和原append参数对应）
obs_concat = pd.concat([obs, pd.DataFrame([aligned_new_obs])], ignore_index=True)
print("obs_concat.shape", obs_concat.shape)  # 输出 (1, 31)

方法2：直接指定列名创建DataFrame #

# 直接用原DataFrame的列名创建单行DataFrame，缺失列自动填充NaN
obs_concat = pd.concat([obs, pd.DataFrame([new_obs], columns=obs.columns)], ignore_index=True)
print("obs_concat.shape", obs_concat.shape)  # 输出 (1, 31)

两种写法都能得到和append完全一致的结果，保证列数为31。

内容的提问来源于stack exchange，提问作者user1245262