Understand Cosine Similarity | 2 Minute Tutorial 

Thank you!

Glad it was helpful!

Thank you!

Great suggestion!

Thank you!

Glad it was helpful!

Glad you liked it!

Glad it was helpful!

thanks for question! Term "learned representation" is usually used when talking about embeddings. Those learned representations (or embeddings) are vectors made by an algorithm, and containing features extracted from input. Similarity or distance measures are used to measure how simmilar/different those embedings are. For example how simmilar are two faces, to images of a car etc.

@@danielkrei Oh, interesting! I thinking that the vector store had all of the embedded vectors that were high dimensional representations of what the model found in terms of similarity or relatedness, and that cosine similarity was the algorithm used to quantify relatedness between two ideas, words, or phrases? I didn't know if there was some new way of representing those connections, if adding to the vector store after the initial embeddings were generated would create issues. Maybe you'd be stuck recalculating the entire "matrix" if you added more info? Thanks for the video!

Isn't the task is to calculate the first one? So which is the most similar to the first one.

In this example I was looking for similarity between the first vector and the other two. Depending on the task you may compute the score for other pairs too.

Good point! It really depends on data you are using but theoretical range is the same as the range of cosine [-1;1]