https://github.com/ansebi/ansnorm_similarity

Determining the distance (dissimilarity) between two vectors: showcase of problems in classic approaches and the solution - AnsNorm distance combining the advantages of euclidean and cosine distances.
https://github.com/ansebi/ansnorm_similarity

Last synced: about 2 months ago
JSON representation

Determining the distance (dissimilarity) between two vectors: showcase of problems in classic approaches and the solution - AnsNorm distance combining the advantages of euclidean and cosine distances.

• Host: GitHub
• URL: https://github.com/ansebi/ansnorm_similarity
• Owner: Ansebi
• Created: 2023-08-25T20:04:38.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2023-08-25T20:06:03.000Z (over 1 year ago)
• Last Synced: 2025-01-21T12:35:56.241Z (4 months ago)
• Language: Jupyter Notebook
• Size: 6.84 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# AnsNorm_Similarity

Determining the distance (dissimilarity) between two vectors: showcase of problems in classic approaches and the solution - AnsNorm distance combining the advantages of euclidean and cosine distances.

Consider computing the distance (non-similarity) between the two given vectors


of length normalized from -1 to 1 in user preference applications.


Distance of 0 between the two must mean that they are completely the same,


larger distances mean less similarity.



Classical methods such as Euclidean distance and Cosine similarity


have their limitations and downsides.


Calculating the dot product as a similarity mesaure is even farther from being ideal.



Euclidean distance does not pay attention to the direction of vectors


while cosine distance ignores the magnitude. Meanwhile dot product as a measure


ignores the common sense.



One possible solution to the problem of evaluating how similar two vectorised personalities are


is AnsNorm Similarity method given below.



Here you can find several cases when the classics aren't ideal


and test ansnorm() function on the same given examples.



FYI:


cosine returns values from 0 to 2


euclidean returns values from 0 to +inf


dot product is any number from -inf to +inf



As for ansnorm,


for input vectors normalised from -1 to 1


it outputs distance values from 0 to 1:


0 for identical entities and 1 for the most dissimilar.