performance optimization when using np.fft.fft on large arrays in NumPy 1.23
I've been researching this but I'm attempting to set up I'm working on a personal project and I'm working with important performance optimization when trying to compute the Fast Fourier Transform (FFT) on large arrays using `np.fft.fft` in NumPy 1.23... The dataset I'm working with consists of a 2D array of shape (10000, 1000). When I run the FFT on this array, I see that it takes an unexpectedly long time to compute. Here's a snippet of my code: ```python import numpy as np # Create a large random 2D array data = np.random.rand(10000, 1000) # Compute the FFT along the last axis fft_result = np.fft.fft(data, axis=1) ``` I initially thought that performing the FFT on each row separately would be manageable, but the computation is taking several seconds. I noticed that when I tried using `np.fft.fft2`, it actually performed even worse, taking nearly double the time. I also experimented with using a smaller sample of my data, say (100, 1000), and the performance was much better, which makes me suspect that the size of the data might be a factor in the FFT computation speed. Is there any specific approach or configuration I can use to optimize the FFT computation on larger arrays? I've also read about using other libraries like CuPy for GPU acceleration, but I haven't switched to it yet. Any suggestions for improving performance while still using NumPy would be greatly appreciated. I'm working on a web app that needs to handle this. For reference, this is a production service. I recently upgraded to Python LTS. Any suggestions would be helpful.