Alternating gradient descent for:
- Hadamard decomposition
- Mixed Hadamard decomposition
Example usage:
from data_utils import makerealdata_full_rank_gaussian
from alternating_gradient_descent import scaled_alternating_gradient_descent_hadDec
D = makerealdata_full_rank_gaussian(250,250,0,1)
D_estimate, [D_1_estimate, D_2_estimate], [A_1, B_1, A_2, B_2], [all_diffs1, all_diffs2] , terminated = scaled_alternating_gradient_descent_hadDec(D, 6, 0.01, 100000)
print(f"Approximation error: {all_diffs2[1]}")Example usage with real (image) data:
import cv2
dataset_name = "rgb_dog"
img = cv2.imread("data/" + dataset_name +".jpg")
D = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
D_estimate, [D_1_estimate, D_2_estimate], [A_1, B_1, A_2, B_2], [all_diffs1, all_diffs2] , terminated = scaled_alternating_gradient_descent_hadDec(D, 20, 0.01, 100000)
plt.figure(figsize=(12, 6))
plt.imshow(D_estimate, cmap='gray'),
plt.xticks([]) # Remove x-axis ticks
plt.yticks([]) # Remove y-axis ticks
plt.tight_layout()
plt.savefig("output/rgb_dog_reconstruction_example.pdf")
plt.show()rgb_dog_reconstruction_example.pdf
The code is tested in Python 3.8.18 and Ubuntu 20.04.2 LTS.
Requirements:
- numpy
- numba
Contact: martino.ciaperoni@aalto.fi