Epoch in Machine Learning

Machine Learning is one of the most used modern technologies in the world. Machine Learning helps human a lot to do their jobs at ease. Epoch is an important concept in Machine Learning. In this article, we will discuss about the epoch in details.

What is epoch in Machine Learning?

Before discussing about epoch let’s understand the basic process to build a model for Machine Learning. Firstly, we have to collect a raw data set from different sources. After collecting the data sets we have to choose one Machine Learning algorithm which will predict our desired things. After choosing the algorithm, we will build our model by passing the data set in that algorithm. In this way, we can build our model. To make the model more accurate, we have to pass the data set many times. But many times Machine Learning developers face problem to pass the data set in Machine Learning model because it takes more memory than allocated memory by the system. So, to solve this problem, we will divide the data set in parts and then we will pass the data set time by time. When the full dataset is passed in the Machine Learning model then it can be called epoch.

Example:

Suppose we have a data set which includes three thousands of data examples then we can divide it into 6 equal parts and then part by part we will pass them through the Machine Learning model. An epoch will be completed when we pass total three thousands of data examples through the model of Machine Learning.

What is batch and batch size?

In the above passage, we have already told that the main problem of passing the whole data set in Machine Learning model is it requires very much memory to do so. For this reason, it is not possible to pass the full data in many cases. So the solution for this problem is batch. Batch is the part of the main data set which we pass through the Machine Learning model at a single time. By passing all the batches, we get epoch. Now, you can face another term which is batch size. It should clear that batch and batch size are totally different things. Batch size is the number of examples which included in a batch.

Example:

If the total data example in the data set is 4000 then we can divide the total dataset into 10 parts. So, there will be 10 batches. The batch size will be total data examples divided by total numbers of batches. So, here batch size will be 400. If we take less batch size such as two hundred then the number of batches increases to 20. If we take greater batch size such as eight hundred then the number of batches will be reduced to 5. So you can see the relation between batch and batch size.

What is iteration?

We have discussed about batch and batch size. Now, let’s understand about iteration. Iteration is a terminology which is used to denote the pass of a batch in the algorithm. When we pass the first batch then the first iteration is completed. Now, you can relate the thing with batch and batch size. The total number of iterations will be equal to the total number batches. 

Example:

Let’s try to understand the concept of iteration by using an example. Suppose, we have the data set which consists of two thousands of data example. It is not possible to pass two thousands data examples at a time. The system can take maximum 1000 data examples. So, we have to arrange the batch and iterations accordingly. If we take batch size of four hundred then the number of batches will be 5. It means we have to pass the different batches one by one. It will happen five times. We know that we can pass one batch in one iteration so there will be total of 5 iterations. After completing five iterations, we will get epoch.

Note:

It is not fixed that you have to pass the data set only once. You can get more than one epoch. It is needed because the Machine Learning model needs more efficiency which cannot be achieved in only one epoch. For this reason we need more than one epoch.