mai-iml's Introduction

Hi there, I'm Victor Badenas 👋

Machine Learning and AI developer, Sound and Music enthusiast.

About Me

💻 I am an AI Tech Lead @ SITA [Nov2023 - Present]
💻 I was an AI Engineer @ Midokura [Feb2022 - Nov2023]
💻 I was an AI Engineer @ Verbio [May2019 - Feb2022]
🌱 I studied Artificial Intelligence and Deep Learning @ FIB-UPC
🌱 I studied Sound and Music Computing @ MTG-UPF
🌱 I studied Telecommunications Engineering @ ETSETB-UPC
💬 Ask me about deep learning, music, image, video games and sound 😊🎵🔊📖
📫 How to reach me: [email protected]📧

💻My Workspace

Packages and Projects

package name	language	description	url	version
frarch	python	Training Framework for Pytorch Experiments	https://pypi.org/project/frarch/

mai-iml's People

Contributors

Watchers

mai-iml's Issues

kNN Index Ordering

In the kNN algorithm, at the __computeKNNIndex function:

    def __computeKNNIndex(self, distanceMatrix):
        knnIndex = [None]*distanceMatrix.shape[0]
        for i in range(distanceMatrix.shape[0]):
            knnIndex[i] = np.argsort(distanceMatrix[i,:])[::-1][:self.k]
        return np.vstack(knnIndex)

If we assume that np.argsort returns the order of the indices in an ordering where the first index has the minimum value, why are reversing this ordering with [::-1]?

For example:

>>> a = [3, 1, 2]
>>> np.argsort(a)
array([1, 2, 0])

I think this is already the correct ordering as the first indexes should represent the minimum values (of distances). If we inverse the ordering, aren't we getting the indexes of the maximum distances?

Recommend Projects