如何在Pandas/Numpy中添加列？一维元组元素Numpy数组扩展方法

Hey there! Let's tackle your two questions one by one, with practical code examples to make things clear.

Pandas #

Pandas makes adding columns straightforward, with a few flexible options:

• Direct Assignment (Simplest Approach)
  Just assign values to a new column name—this modifies the DataFrame in-place:

  import pandas as pd
  
  df = pd.DataFrame({'A': [1, 2, 3], 'B': [4, 5, 6]})
  # Add a column with a constant value
  df['C'] = 7
  # Add a column calculated from existing columns
  df['D'] = df['A'] + df['B']
  

• Using assign() (Chainable, Non-Destructive)
  If you want to avoid modifying the original DataFrame and prefer chained operations, use assign():

  # Create a new DataFrame with additional columns
  new_df = df.assign(E=lambda x: x['A'] * 2, F=[10, 20, 30])
  

• Insert at a Specific Position
  Use insert() to place the new column at a specific index:

  # Insert column 'G' at position 1 (between 'A' and 'B')
  df.insert(1, 'G', [100, 200, 300])
  

NumPy #

NumPy arrays are fixed-size, so adding a column means creating a new array. Here are common methods:

• Using np.hstack() (Horizontal Stack)
  Combine the original array with a new 2D column array:

  import numpy as np
  
  arr = np.array([[1, 2], [3, 4], [5, 6]])
  # Create a new column (must be shape (n, 1))
  new_col = np.array([[7], [8], [9]])
  # Add the column to the original array
  new_arr = np.hstack((arr, new_col))
  

• Using np.concatenate()
  Similar to hstack, explicitly specify axis=1 to stack columns:

  new_arr = np.concatenate((arr, new_col), axis=1)
  

• For 1D Arrays
  Reshape the 1D array to 2D first before adding columns:

  arr_1d = np.array([1, 2, 3])
  # Reshape to (3,1) and combine with a new column
  new_arr = np.hstack((arr_1d.reshape(-1, 1), np.array([[4], [5], [6]])))
  

Since tuples are immutable, we can't modify them in-place—we need to create new tuples. Here are efficient ways to keep the array shape as (n,):

• List Comprehension (Fast & Readable)
  This is the most straightforward method for this use case:

  # Original array with tuple elements (dtype=object is key here)
  arr = np.array([(1, 2), (3, 4), (5, 6)], dtype=object)
  # Add an element (e.g., 99) to each tuple
  new_arr = np.array([t + (99,) for t in arr], dtype=object)
  # Verify shape: new_arr.shape will be (3,)
  

• Using map()
  A concise alternative if you prefer functional-style code:

  new_arr = np.array(list(map(lambda t: t + (99,), arr)), dtype=object)
  

• Note on np.vectorize()
  While you can use np.vectorize, it’s essentially a wrapper around a loop and won’t outperform list comprehensions here. Example:

  def add_element(t):
      return t + (99,)
  
  vectorized_add = np.vectorize(add_element)
  new_arr = vectorized_add(arr)
  

  Keep in mind that object-dtype arrays (storing tuples) don’t get the full vectorization speed of numeric arrays, but list comprehensions are still efficient for most practical cases.

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