Exam AWS Certified Machine Learning - Specialty topic 1 question 224 discussion

to be able to find similar data points for each test data point.

K-means unsupervised learning

It is A Memory Efficiency: K-nearest neighbors (k-NN) doesn't require storing a model with learned parameters, as it's an instance-based learning algorithm. It simply memorizes the training dataset. Therefore, it saves on memory costs compared to models with learned parameters like linear learners. Dimension Reduction: By employing dimension reduction techniques like Principal Component Analysis (PCA) in conjunction with k-NN, you can reduce the dimensionality of the dataset, which helps in saving memory costs. This makes k-NN with dimension reduction a suitable choice when memory efficiency is a concern. Similar Data Points: K-nearest neighbors naturally provides a measure of similarity between data points. Given a test data point, it finds the k nearest neighbors in the training data. This fulfills the requirement of being able to find similar data points for each test data point.

A. K-nearest neighbors (k-NN) with dimension reduction ( KNN - Useful for Classification task for different types of veggies based on features + Dimensionality reductions like PCA can be applied prior to KNN to reduce no.of features in dataset , thereby saving memory costs during training and model deployment - also remove noise and data redundancy )

https://www.linkedin.com/advice/3/what-difference-between-knn-k-means-skills-computer-science-cx1hc

This is un unsupervised clustering problem not a classification one (A). k-means is a better choice from memory efficiency perspective.

While A is most voted comment, but knn is really high on memory usage as it stores the data points information to make predictions. Just voting for it because it mentions dimensionality reduction is obtuse. C is the next most probable candidate that fits the bill on every account.

Should be A, as only A is can be used for classification, finding similar data points and dimensionality reduction

A. YES - K-NN will find the similar datapoints, and dimension reduction will save memory B. NO - Linear learner is for regression or classification, not finding similar data points C. NO - K-means is for unsupervised clustering, not find closest data ponits D. NO - Principal component analysis (PCA) with the algorithm mode set to random

C doesn't solve the "too many features" problem + It's well defined the vegetable classes. A is the way

KNN to reduce dimensionality which may help reduce memory utilisation.

It's A, needs to reduct the dimensionality of the dataset.

They wanna less feature

I will go with A. C is not valid, as K-means is a clustering algorithm that can group similar data points together. However, it does not perform classification, and it is not clear how it addresses the memory cost and similarity search requirements mentioned in the question.

It should be C, because it is unsupervised classification problem.

option A suggests using the k-nearest neighbors (k-NN) algorithm with dimension reduction. The k-NN algorithm can be used for classification tasks and dimension reduction can help reduce memory costs. Additionally, k-NN can be used for finding similar data points. K-NN is a simple algorithm that works well with high-dimensional data and can find similar data points.

https://ealizadeh.com/blog/knn-and-kmeans/#:~:text=unsupervised%20learning%20algorithm.-,K%20in%20K%2DMeans%20refers%20to%20the%20number%20of%20clusters,using%20different%20values%20for%20K. KNN does use lot of memory because in lazy learning it stores (memorizes) the training dataset. AWS sagemaker however has an improved version of this algorithm. Because the questions does not mention we have labels, we cannot use supervised learning K-means : unsupervised and helps us to classify different vegetables based on their many features. This also find "similar data points for each test data point" c