Top-K and Relative Accuracy

# Get top-K accuracy from results. Given a query and its related results, it'll find if any of the Ground Truth are in results.
# Or How many of the results are in Ground Truth

import numpy as np

def get_relative_acc(test_result_indices:[np.ndarray,list],similar_image_indices:[np.ndarray,list])->float:
    '''
    Check HOW MANY similar images are in returned results for single test image
    '''
    # because then only it'll be relative. If an image has 2,3,4 similar images, it won't matter. It'll normalize relative acc
    divide_by = min(len(test_result_indices),len(similar_image_indices)) 
    return len(set(similar_image_indices).intersection(set(test_result_indices)))/divide_by


def get_top_k_acc(test_result_indices:[np.ndarray,list],similar_image_indices:[np.ndarray,list])->int:
    '''
    Check if ATLEAST 1 similar image is in results. K = len(test_result_indices)
    '''
    for i in similar_image_indices:
        if i in test_result_indices: # if there exists atleast one single common image, return True
            return 1  
    return 0 # if there are no common images in 5 results, return False


#---------------------------------------------#
# Case 1
test_result_indices = [1,3,5,7,11,13]
similar_image_indices = [1,2,3,4,5,6,7,8,9]

print(get_relative_acc(test_result_indices,similar_image_indices))
print(get_top_k_acc(test_result_indices,similar_image_indices))

# Case 2
test_result_indices = [1,2,3,4,5,6,7,8,9]
similar_image_indices = [1,3,11,13]

print(get_relative_acc(test_result_indices,similar_image_indices))
print(get_top_k_acc(test_result_indices,similar_image_indices))

#---------------------------------------------------------------#



# Check for all images
# acc = []
# for image in test_images: # get the indices of its similar image
#     top_5_result_indices = apply_cosine_get_indices(image)
#     result = calculate_relative_acc(top_5_result_indices,similar_indices)
#     acc.append(result)
# print(sum(acc)/len(acc))