从R转Python：NumPy中实现类似Merge操作的高效方法咨询

Hey there! I totally get the confusion when switching from R's merge to NumPy—concatenate just stacks arrays end-to-end, which isn't what you want when you need to match rows by an ID key. Let's walk through how to replicate R-style merging with NumPy, plus a bonus pandas method that might make your life easier (since it’s built for this exact kind of tabular data work).

First, let's recap your datasets #

import numpy as np

d1 = np.array([['1a2', '0'], ['2dd', '0'], ['z83', '1'], ['fz3', '0']])  # ID, Label
d2 = np.array([['1a2', '33.3', '22.2'], ['43m', '66.6', '66.6'], ['z83', '12.2', '22.1']])  # ID, val1, val2

Option 1: NumPy Implementation #

NumPy doesn’t have a built-in merge function, but we can build it using array indexing and set operations.

Inner Join (default behavior of R's merge) #

This keeps only rows where the ID exists in both datasets, just like merge(d1, d2, by="ID") in R.

1. Extract the ID columns from both arrays:

   ids1 = d1[:, 0]
   ids2 = d2[:, 0]
   

2. Find IDs that exist in both arrays:

   common_ids = np.intersect1d(ids1, ids2)
   

3. Get the indices of these common IDs in each array, making sure they’re ordered to match:

   # Indices in d1 for common IDs
   idx1 = np.where(np.isin(ids1, common_ids))[0]
   # Match the order of d1's common IDs to get corresponding indices in d2
   idx2 = np.array([np.where(ids2 == id_)[0][0] for id_ in ids1[idx1]])
   

4. Stack the matching rows (we exclude d2's ID column since it’s duplicated):

   merged_inner = np.hstack([d1[idx1], d2[idx2, 1:]])
   

   The result will look like this:

   array([['1a2', '0', '33.3', '22.2'],
          ['z83', '1', '12.2', '22.1']], dtype='<U3')
   

Left Join (keep all rows from d1, match d2 where possible) #

This is equivalent to merge(d1, d2, by="ID", all.x=TRUE) in R. We’ll initialize the merged array with NaN for missing matches:

# Create a base array with d1's data + empty columns for d2's values (filled with NaN)
merged_left = np.hstack([d1, np.full((d1.shape[0], d2.shape[1]-1), np.nan, dtype=object)])

# Loop through each ID in d1 and fill in matching values from d2
for i, id_ in enumerate(ids1):
    match_idx = np.where(ids2 == id_)[0]
    if len(match_idx) > 0:
        merged_left[i, 2:] = d2[match_idx[0], 1:]

Result:

array([['1a2', '0', '33.3', '22.2'],
       ['2dd', '0', nan, nan],
       ['z83', '1', '12.2', '22.1'],
       ['fz3', '0', nan, nan]], dtype=object)

Option 2: Pandas (Way Simpler for Tabular Data) #

If you’re open to using pandas (which is the Python equivalent of R’s data frames for most tasks), the syntax is almost identical to R’s merge:

import pandas as pd

# Convert NumPy arrays to DataFrames with column names
df1 = pd.DataFrame(d1, columns=["ID", "Label"])
df2 = pd.DataFrame(d2, columns=["ID", "val1", "val2"])

# Inner join (default)
merged_inner_pd = pd.merge(df1, df2, on="ID")

# Left join
merged_left_pd = pd.merge(df1, df2, on="ID", how="left")

This gives you the exact same results as the NumPy method, but with way less code—perfect if you’re transitioning from R and want a familiar workflow.

Quick Note #

NumPy is great for numerical computations, but it’s not designed for structured, labeled data like this. Pandas is the go-to tool for this kind of merging and data manipulation in Python, so it’s worth learning if you’re doing a lot of R-to-Python data work.

内容的提问来源于stack exchange，提问作者Jibril