What really machine learning is those 3 magical word

Subject:What if i say Machine Learning and Deep Learning are the things that have only 3 principles called:

Photo by Arnaud Mariat on Unsplash

Hello World Welcome to my simple Machine Learning tutorial , so i’m sanjay recently learned machine learning from various sources

1) So Where I learned ML:

I have searched for various Sources even downloaded some paid courses syllabus also , but that not worked because i didn’t found anything useful(useful means in the sense of simplified math which means math parallel to ML ) , suddenly on edX i found this course that’s very useful from UCSanDiegoX and udacity Machine Learning course are the best that i think , because i even took Andrew ng ML courses ALSO.

1. edX Machine Learning course
2. udacity Machine Learning course

[P9-DS] My Data Science Path 2019 -NOV 8th week

2) The 3 magical Principles :

1. Tensors
2. Loss Function
3. Distance metrics

Agree with me these are things that used to build/create any type of new machine leaning algorithm even Deep Learning algorithm also , I don’t know about(reinforcement Learning ). Let me explain you in detail.

2.1) Tensors :

No matter what kind of data(image,text,numbers) you have to use with ML Algorithms(SVM,Naive Bayes,KMean,KNN,,Gaussian,Boosting,Bagging , Neural Networks, etc) ,the data must be converted to Vector / Array / Tensor[ numpy array / torch tensor , some other tensor like format].

Linear Algebra Comes into Action specifically [matrix and vectors and it’s operation ] will play major role in it.

2.2) Loss Function:

Loss Function and Optimizer will play the major role in the process of training a Model(any ML algorithm) this compute how loss / steep need to take how much weight needed to apply to a tensor in order to reduce the loss.

Linear Algebra [matrix multiplication] and calculus [Slope finding] will play the math role in this part called “LOSS FUNCTION and OPTIMIZER”.

Here loss function = predicted value — original data point.

weight needed to apply but where right?

When you learn ML Algorithms (all algorithms except Probability models) you will find something called :

M.X+b => Equation of line

W.X+b => Equation of Hyperplane

w => some weight

X =>Data Point -> Tensors

b => some constant

the optimizer will optimize or try to reduce W it’s a convex loss function else it will find a global minima.

2.3) Distance metrics:

This distance metrics is used to calculate the distance between data points that are scattered in all the dimensional , some people this that distance metrics are used in Clustering , but according to me this is the most important function or most important parameter in loss function.

so which part math is used to calculate the distance between again Linear Algebra [norms]:

1. 1-norm
2. 2-norm(Euclidean distance/Euclidean norm)
3. P-norm
4. infinity norm

choosing right norm/distance function will increase the accuracy .

3)What about Kernels and PCA?

I’m not talking about that here may be in later because they are super simple concept.

[P8-DS] My Data Science Path 2019 OCT — 7th Week

[P7-DS] My Data Science Path 2019 OCT — 6th week