In this post on Getting Familiar with Machine Learning, I will explain the basic concepts of Machine Learning. Basically, Machine Learning refers to a set of techniques that when imparted in a system allows learning from the data. Moreover, learning improves with time and with more data. Hence, we can say that the system learns from the experience even when it is not instructed to do so explicitly. Therefore, Machine Learning Systems get valuable information or insight from the input data.
Classifying Machine Learning Techniques
We can broadly classify Machine Learning techniques into three major categories.
- Supervised Learning
- Unsupervised Learning
- Reinforcement Learning
When the input data has labels, we can use the Supervised Learning. Usually, the input data is available as a .csv file containing the rows and columns where the columns denote features of the data and rows denote the individual records. One of the columns contains the value of the label. The machine learning technique tries to predict the value of the label given the values of features. The label may contain distinct values or they may contain the continuous numeric value. In the latter case, it is called a regression problem. However, if the label contains discrete values, then it is called a classification problem.
So, in the case of supervised learning, the model training takes place prior to predicting. We train our supervised learning model using the historical data which already has labels on it. The model training takes several iterations and once training completes, our model is ready to make predictions on new data set.
Sometimes, we don’t have historical data with input labels. In that case, a machine learning technique tries to find the patterns present in the data. Basically, similar data values are grouped together to form a cluster. In this case, the learning takes place without labels and different clusters are formed. The formation of clusters containing similar data points is termed as Clustering. We refer to this type of Machine Learning technique as the Unsupervised Machine Learning technique.
In fact, Clustering tells nothing about the classes or labels that each cluster indicates. It is up to the experts to decide the class label for each cluster.
Learning also takes place through trial and error and each action may result in either a penalty or a reward. Basically, the system learns which of the actions results in the greatest reward. We refer to this kind of learning as Reinforcement learning.
The application areas of Reinforcement learning are self-driving cars, learning through camera feeds, or learning to play a video game. As far as, the evaluation of Reinforcement Learning is concerned, it is much easier in comparison to Supervised and Unsupervised Learning Techniques. In fact, the model evaluation is built into the model training and depends upon how well the system has performed the actual task assigned to it.
To summarize, we classify Machine Learning Techniques broadly in three categories – Supervised Learning, Unsupervised Learning, and Reinforcement Learning. In essence, it is the application and the availability of data that decide which of the above technique is more suitable.