Unsupervised Machine Learning Algorithms Implementation From Scratch

Following are the implementations of Unsupervised machine learning algorithms from scratch without using any Machine learning libraries

K-Means clustering GitHub Link

Implementation of K-means clustering from scratch using python. The implementation includes the two steps process

  1. assigning all samples to one of the closest centroids
  2. move centroids to the middle of those assigned samples.

However, the implementation is a little tricky for that I’ve implemented the following methods

  • findClosestCentroids
  • computeCentroids
  • runkMeans

Anomaly Detection GitHub Link

Implementation of Anomaly Detection from scratch using python.

Step 1: To perform anomaly detection, first, need to fit a model to the data’s distribution

Given a training set {x(1), …, x(m)} (where x(i) ∈ Rn), you want to estimate the Gaussian distribution for each of the features xi .

For each feature i = 1 . . . n, you need to find parameters µi and σ2i that fit the data in the i-th dimension x(1)i,…,x(m)i (the i-th dimension of each example).

The Gaussian distribution is given by

p(x;μ,σ2)=12∗π∗σ√exp(−(x−μ)22σ2)

where µ is the mean and σ2 controls the variance.

Step 2: Estimating parameters for a GaussianYou can estimate the parameters, (µi, σ2i), of the i-th feature by using the following equations.

To estimate the mean, you will use: μi=1m∑mj=1x(j)i σ2=1m∑mj=1(x(i)−μ)2

Step 3: Selecting the threshold, ε

Now that you have estimated the Gaussian parameters, you can investigate which examples have a very high probability given this distribution and which examples have a very low probability.

The low probability examples are more likely to be the anomalies in our dataset. One way to determine which examples are anomalies is to select a threshold based on a cross-validation set. and here I have done it by running my algorithm for different values of ε to select the threshold ε using the F1 score on a cross-validation set.

Principal Component Analysis(PCA) GitHub Link

The motivation of principal component analysis is mainly about data compression – for example, how to casting a high dimension data down to a lower dimension space. This will help saving space of data storage, also improving certain algorithm speed, and also to help visualization since a data more than 3D, 4D will become impossible to be visualized directly.

Courses & Certificates

Conclusion

The following link to my Programming showcase repository on Github includes implementation of every algorithm discussed here