How to Efficiently Find Row Indices of Specific Values in a NumPy Array?

Finding the Index of Specific Values ​​in a NumPy Array

Problem:

Given an array X, determine the row indices of specified values ​​stored in searched_values. For instance, in the following example, we seek the indices for the values ​​[4, 2], [3, 3], and [5, 6] within the array X.

X = np.array([[4,  2],
              [9,  3],
              [8,  5],
              [3,  3],
              [5,  6]])

searched_values = np.array([[4, 2],
                            [3, 3],
                            [5, 6]])

The desired output would be:

[0, 3, 4]

Solutions:

Approach 1: NumPy Broadcasting

is a way to utilize np.where and the broadcasting functions are as follows.

np.where((X == searched_values[:, None]).all(-1))[1]

Approach 2: Memory-Efficient Linear Index Conversion

How to convert each row to a unique linear index using NumPy's np.ravel_multi_index There is.

dims = X.max(0) + 1
out = np.where(np.in1d(np.ravel_multi_index(X.T, dims),
                       np.ravel_multi_index(searched_values.T, dims)))[0]

Approach 3: Memory-Efficient and Optimized Linear Index Conversion

NumPy's np.searchsorted can also be used to quickly find linear indexes. can.

dims = X.max(0) + 1
X1D = np.ravel_multi_index(X.T, dims)
searched_valuesID = np.ravel_multi_index(searched_values.T, dims)
sidx = X1D.argsort()
out = sidx[np.searchsorted(X1D, searched_valuesID, sorter=sidx)]

How np.ravel_multi_index works:

np.ravel_multi_index creates a linear index representation of a multidimensional index. Interpret each row as an index into an n-dimensional multidimensional array and generate the corresponding linear index.

For example, applying np.ravel_multi_index(X.T, dims) to a sample array X:

np.ravel_multi_index(X.T, dims)
# Output: array([30, 66, 61, 24, 41])

This represents the linear index that each row of array X has. This linear index can be used to uniquely identify each row in the array.

The sample code corresponding to each method is below.

# Approach 1
print(np.where((X == searched_values[:, None]).all(-1))[1])

# Approach 2
dims = X.max(0) + 1
print(np.where(np.in1d(np.ravel_multi_index(X.T, dims),
                       np.ravel_multi_index(searched_values.T, dims)))[0])

# Approach 3
dims = X.max(0) + 1
X1D = np.ravel_multi_index(X.T, dims)
searched_valuesID = np.ravel_multi_index(searched_values.T, dims)
sidx = X1D.argsort()
print(sidx[np.searchsorted(X1D, searched_valuesID, sorter=sidx)])

You can easily determine the row index of a particular value in array X using either approach.

The above is the detailed content of How to Efficiently Find Row Indices of Specific Values in a NumPy Array?. For more information, please follow other related articles on the PHP Chinese website!