如何基于numpy数组rel的抽取结果生成对应eta_extract数组？

First, let's break down the core relationship we're working with:

• Every element in eta maps one-to-one with the positions of 0s in the original rel array, following the order those 0s appear in rel.
• rel_extract is a subset of rel elements, so we need to trace the 0s in rel_extract back to their original positions in rel, then link those positions to the corresponding values in eta.

Critical Note #

To make this mapping accurate, you must know the indices from the original rel that were used to create rel_extract (let's call this extract_indices). Without this, we can't definitively match 0s in rel_extract to their eta values—since multiple 0s in rel look identical in the subset.

Step-by-Step Implementation #

Let's use your example data to walk through the code:

import numpy as np

# Original datasets
rel = np.array([1,0,0,1,1,0,1,0,1,0,0])
eta = np.array([2,3,10,16,4,3])
rel_extract = np.array([1,0,0,1,0])
# Replace this with your actual extraction indices (the positions pulled from rel to make rel_extract)
extract_indices = np.array([0,1,2,3,9])  # This gives rel[extract_indices] = rel_extract

1. Capture all positions of 0s in the original rel
   This creates an ordered list where each index directly corresponds to an element in eta:

   zero_indices = np.where(rel == 0)[0]
   # Result: array([ 1,  2,  5,  7,  9, 10])
   

2. Find original rel indices that map to 0s in rel_extract
   Filter extract_indices to keep only positions where rel_extract has a 0:

   zero_extract_original_indices = extract_indices[rel_extract == 0]
   # Result: array([1, 2, 9])
   

3. Map these original indices to their positions in zero_indices
   Since zero_indices is sorted (from np.where), we can use np.searchsorted for fast lookup:

   eta_positions = np.searchsorted(zero_indices, zero_extract_original_indices)
   # Result: array([0, 1, 4])
   

4. Generate eta_extract by indexing into eta

   eta_extract = eta[eta_positions]
   # Final result: array([2, 3, 3])
   

Concise One-Liner Version #

You can combine steps for brevity without losing clarity:

zero_indices = np.where(rel == 0)[0]
eta_extract = eta[np.searchsorted(zero_indices, extract_indices[rel_extract == 0])]

What If You Don't Have extract_indices? #

If you didn't record the indices used to create rel_extract, you'll need to infer them by matching element sequences—but this is unreliable if rel has repeated patterns. For example, you could use np.where to find all positions in rel that match the rel_extract sequence, but this only works for unique or contiguous subsets. Always track extract_indices when creating rel_extract for best results!

内容的提问来源于stack exchange，提问作者user23299