Machine learning is a buzz word these days. Let us just try to understand from a beginner point of view in layman language.
Machine learning, as its name suggests, is a method of teaching machine to learn themselves from given and future data. You give your algorithm some initial data and ask it to predict in future. It is completely inspired from human beings. Like the human, as they gain experience, they become proficient in taking decisions. Similarly, with time machine learning algorithms becomes better at decision making. If you have noticed on Facebook, it gives you friend suggestions "People you may know / Video you may like". It never says people "you should know, video you should/will like" because machine learning algorithms work on probabilistic model. If you have noticed, with time or more you use Facebook, it gives you better and more confined results.
At the first thought, it seems a very challenging and fascinating task to devise machine learning algorithms but basic idea behind them is not as complicated as it seems at first thought.
let us say you want to devise a machine learning algorithm to predict price of house. you have been given data of 10 houses. price and all details like area,number of rooms, locality where it is present, furnished/unfurnished and I ask you to predict price of another 2 houses based on this data.
What would you do. You would try to find trend in price w.r.t. various parameters. e.g. price per square meter, price w.r.t. to number of rooms,etc
Say two houses have all the same parameters except locality and it has same price. it may be true to say that location does not affect on price of house. but this is also possible that data is not sufficient. we can make atleast some prediction based on this observation.
How would a machine predict? in simplest model of machine learning, it will assume linear model.
A,B,C &D are weight of various parameters. If some parameter does not contribute to price of house, its corresponding weight value will be zero. A0 is bias or in simple words you can say this is the base price which is independent of any parameter.
job of algorithm is to find optimal values of A0 , A. B, C and D. You have to find these parameters such that this equations fits for given data and it gives a good prediction for future data.
in the simplest approach, we can minimize the error :
we have used here the squared error function.
Once weight vectors are calculated, your model is ready for predictions. You just have to put different input data in this equation, you will get approximated price of house.
Isn't it really cool and easy :) I never knew in my 11th- 12th standard that what I was studying will be used for such a great technology in future.
Machine learning, as its name suggests, is a method of teaching machine to learn themselves from given and future data. You give your algorithm some initial data and ask it to predict in future. It is completely inspired from human beings. Like the human, as they gain experience, they become proficient in taking decisions. Similarly, with time machine learning algorithms becomes better at decision making. If you have noticed on Facebook, it gives you friend suggestions "People you may know / Video you may like". It never says people "you should know, video you should/will like" because machine learning algorithms work on probabilistic model. If you have noticed, with time or more you use Facebook, it gives you better and more confined results.
At the first thought, it seems a very challenging and fascinating task to devise machine learning algorithms but basic idea behind them is not as complicated as it seems at first thought.
let us say you want to devise a machine learning algorithm to predict price of house. you have been given data of 10 houses. price and all details like area,number of rooms, locality where it is present, furnished/unfurnished and I ask you to predict price of another 2 houses based on this data.
What would you do. You would try to find trend in price w.r.t. various parameters. e.g. price per square meter, price w.r.t. to number of rooms,etc
Say two houses have all the same parameters except locality and it has same price. it may be true to say that location does not affect on price of house. but this is also possible that data is not sufficient. we can make atleast some prediction based on this observation.
How would a machine predict? in simplest model of machine learning, it will assume linear model.
Price of house, P = basic property rate (A0) + A * number of rooms + B * area of house + C * floor on which it is located + D * locality
job of algorithm is to find optimal values of A0 , A. B, C and D. You have to find these parameters such that this equations fits for given data and it gives a good prediction for future data.
in the simplest approach, we can minimize the error :
Error in prediction, E = (real price of house - predicted price of house)^2
we have used here the squared error function.
Once weight vectors are calculated, your model is ready for predictions. You just have to put different input data in this equation, you will get approximated price of house.
Isn't it really cool and easy :) I never knew in my 11th- 12th standard that what I was studying will be used for such a great technology in future.
Good one
ReplyDelete