Machine learning

Day 1

Machines work from our instructions. The term “Machine learning” comes when they start to work from their own experience just like humans try to learn from their experience. Tesla’s self-driving car, apples Siri, microsoft’s Cortona, google now, Sofia AI robot etc. are some examples of machine learning applications. Like microsoft’s cortona, siri continuously learn about their users. Such that they can anticipate the users needs.

So, basically the ML is about focusing on designing a system which can learn and make predictions based on the machines past experiences/ data. Instead of using the programs beforehand to carry out a action, based on its experience, it takes the necessary steps.

How does the machine learning work?

Objective- To categorize underweight and overweight people. Let the red dots present underweight and blue dots present overweight persons.

If someone wants to ask the person B’s weight lies in which category. We can easily answer that by saying category B. What if someone asks what is the status of point A which is sort of in between underweight and overweight category. Here machine learning comes into the picture. A circle around that point can be drawn and it is seen that four red points lies in the circle. Hence, we can categorize it within underweight category. This is how the machine learning algorithm works. It learns from the data, develops prediction model and predicts the status of new data points. This is one example of a machine learning algorithm.

(Algorithms are methods or procedures taken in other to solve a problem, while Models are the output of the algorithms. For example, we want to find the value of y for a certain value of x in when they are linearly related.

First, we use the algorithm that shows the linear relationship between x and y, which is

y = mx + c

Then, we create a model for a certain case where m = 5 and intercept c = 6, then we will have

y = 5x + 6

with the developed model we can find the value of y for different value of x.)

Things that I learned

1. What is machine learning?
2. Real life applications of machine learning
3. How machine learning works through an example.
4. Difference between algorithm and model in machine learning.

Day 2

Machine learning types

1. Supervised learning
2. Unsupervised learning
3. Reinforcement

Supervised learning

 Why it is named so? This type of learning is same as when we are solving problems under the guidance of our teacher or we are supervised by the teacher. We continue with the learning process till the teacher is happy with our performance level. The process of supervised learning is exactly like that only. The data are labeled (labeled means some meaningful information and tag is attached to the data) and are divided into two sets training set and testing set. With the appropriate algorithm, a model is developed by the machine and some percent of data is used to train the set of data and the other portion of data is used to test the set of data, till the model achieves a required accuracy level. For example, we want the model to distinguish whether the apple is red or green. We can feed the data of the apples color; the model will use some portion of data to train itself whether it is green or red. Then, with the other portion of data, it will test if it has achieved the require accuracy level or not. If it is successful in predicting the apple color, then it is good to go.

Supervised learning can be divided into two types of learning problem

1. Classification (when the output is categorical data like color of apple is red or green)
2. Regression (when the output is continuous data like what will be the revenue of the project)

Some real-life applications of supervised learning- Cortona, or siri trains itself with your voice and do the work as you have commanded. (Speech pattern recognition-Classification)

Next is the biometric attendance, when we train the machines, after a few inputs the machine recognizes your thumb pattern and can identify you in future. (Classification)

In health sectors, based on the patient’s health attributes, the readmission rate is calculated. (Regression model)

Unsupervised learning

Why is it named so? Unlike supervised learning, it does not have any teacher or any labeled data that means it has only input data and no out put data. So how does it work? It works by finding the pattern in the data without any human intervention.  One example of unsupervised learning is clustering. For example there are fruits and the machine will classify according to the data with it but it cannot add label to those data.

Another example is when you watch a cricket match for the first time. You will classiy them may on the basis of their jersey or the batsman or bowlers etc. It means without any prior knowledge, you will try find a pattern in those data.

Real life applications- In health care industry it is used to classify MRI data normal or abnormal on the basis of different images or features.

 In banking sectors, it helps in classifying the characteristics on the basis of behavioral characteristics to form segments.

Things that I have learned-

1.  Types of Machine learning
2. What is supervised learning and its application?
3. What are labeled data?
4. What are the main types of supervised learning?
5. What are some real life applications of supervised learning?
6. What is unsupervised learning?
7. What are some real life applications of unsupervised learning?

Day 3

Reinforcement learning

• It is a technique in which the machine learns by trial-and-error method with the feedback from its own experience. The programmer set rewards for all the correct actions that the machine performs and gives a penalty for all the wrong work. Here the goal of the machine is to maximize the reward. In short the programmer gives the reward or penalty for achieving the goal but does not tell how to reach the goal.

Practical application includes online recommendation, in banking sector it is used to get the best offer model based on the user’s acceptance or rejection of the offer.

Steps to solve a ML problem

1. Collection of data (supervised learning requires labeled data) More the data better is the result. Depending the quantity and quality of the data model accuracy can be improved.
2. Data preprocessing or cleaning of data (like missing data, outliers, wrongly labeled data should be fixed)
3. Feature engineering (representing data with features, like for email’s language English-0 and hindi-1). The algorithm can learn better with more features
4. Algorithm selection and training (Supervised learning algorithms- linear regression, Naïve Bayes, Support vector machine, Decision tree, random forest, k nearest neighbor) Unsupervised learning algorithms- K means clustering, KNN, Hierarchical clustering, gaussian mixture model) Reinforcement learning algorithm- Q- learning)
5. Making prediction

Things that I have learned

1. Reinforcement learning definition
2. Real life application of reinforcement learning
3. Steps to solve a machine learning problem
4. Types of algorithm of this 3 types of learning