 Welcome back to this video session. So this video session titled is machine learning terminologies and life cycle. So let us see what has been planned for this video session. I mean after watching this video the audience would be able to describe machine learning life cycle in detail. So this video would be introducing you the machine learning terminologies and then we will be discussing what exactly is the process and what exactly is the steps involved in machine learning life cycle. So let us start with our machine learning terminologies. So when we are learning about machine learning it is very important to know what all are the basic terminologies involved in machine learning. So let us look at the terminologies one by one. So the fundamental terminology which we always start with in any of the machine learning discussions or any of the machine learning architectures is we always very frequently talk about a word called as data sets. So what exactly do we mean by data sets? Data sets in machine learning is nothing but a well organized and structured or it could be in some cases an unstructured as well is a set of records or instances which we call as a data upon which my machine learning model would be actually be learning some insightful patterns to predict an outcome for a future event. So when we talk about data sets in a layman words I can explain it as is nothing but a set of data or a data collections upon which my machine learning model would be learning on. So further going on when we go beyond the concept of data set we usually very frequently encounter the terms called as features and labels. So if data collection is termed as a data sets then the individual further description of data set is always in the form of features and labels. So features and labels are nothing but we can visualize these terms as for example let us consider a small excel data set wherein we have some rows and columns usually the columns are form what right now we are calling them as features and labels. So we have an example in a next slide where we will be exactly saying an example of what exactly are the features and what exactly are the labels. So in general each row in a data set is always form from labels and features. So going beyond we do have one more term called as model and in general we call them as machine learning model. So what exactly does a machine learning model mean? So here when I say model and machine learning it is nothing but in the computational entity and when I am saying a computational entity it could be an algorithm or a step by step process or any a computational paradigmic entity which describes a relation between features and labels. So when we talk about features and labels it is nothing but labels are those columns in your data set whose value your ML model tries to predict given the feature values for that consecutive or concerned rows. So in short model is nothing but it is a computational entity which describes a relationship between a features and labels. So given a few set of features we can easily predict what could be the possible outcome for a label. So going ahead we have few more terms called as parameters and hyperparameters. So talking about parameters and hyperparameters those are still computational variables but an exact meaning of parameters and hyperparameters can be inferred by visiting back the definition of features and labels. Parameters and labels are the input data which those are the actual part of data set but when your ML model is trying to form or find relationship between that it has to even reliance some additional variables and these variables are something which are not a part of data set but these are something which are computed during the process of training or during the process of machine learning process of building machine learning model. So parameters are those whose variables are learned from features and features when your ML algorithms are in a process of trying to learn the relation between features and labels and hyperparameters are those parameters in this process of learning a model whose values by far remain constant. So now since we are at the very introductory slide I am avoiding taking an example but if I am required to take an example of hyperparameter there is an algorithm called as k-means which does a clustering. So in this the k defines the number of clusters we want my algorithm to produce here the value of k can be termed as hyperparameters. At the same time in the same algorithm when we talk about the centroid distance radius which is internally computed based on the values fed to that algorithm those fall under the category of parameters. So let us move ahead what other is the training data set and validation data set. These terms make very often appearance when we are talking about a supervised machine learning which we will be seeing in a further slides. So whenever we have a data set my model or my algorithm or my ML learning process doesn't work on entire data set. Generally this data set is split into two parts. First part is called as training set upon which my ML order would be actually be learning how features and labels are interlinked and how I can predict a value for the label. So once the ML model learns the relationship we test the accuracy or the credibility of that learning process by actually passing the input from the validation set. The validation set is that part of data set through which your ML model would never go through during the learning process. Only upon when we find that the ML has learned completely we test to what extent my machine learning model has learned by asking machine learning to predict the values for the validation data set. And when we are talking about that sometimes we get too confused we use train and testing data set and now you might ask me what exactly is the validation data set. So when during the process of learning even though we have been working on a training set a some part of training set is further removed out to fine tune the learning process. So when I am saying fine tune the learning process I am talking about making some trial and error like experiment runs to make sure my machine learning model is learning fine. So in general your data set is always split into three parts the train data set the test data set and validation data set during machine learning process the data through which your ML model goes through is usually the train data set and testing data set based on which my ML model learns and then there is an actual accuracy of model is always determined by running the model on validation data set. See what exactly is a machine learning life cycle. So when we talk about machine learning life cycle what we are talking about is a given a problem and you wish to solve the problem by using machine learning these are the steps which you need to follow. So these ensure that the steps required in machine learning are properly done in a sequence which would lead to a fruitful outcome by machine learning process. Let us see those steps one by one and we will be describing each step in detail. So the first one for any machine learning is to identify the data. So without data we cannot frankly speaking without data there is no machine learning. So the very important step in an entire machine learning life cycle is to first make sure on what data you are trying to put your machine learning model to a practice through which your machine learning model will be learning the data. So more the good data more the rich data better the machine learning model or the better the capability of machine learning model to predict an outcome. Second make sure once you identify the data the data is prepared properly. For example by if you go through the my earlier slide we were able to see the table where you could see the data was properly structured in column and rows and format. Usually more structured the data better than my computational algorithms and it in indirectly or directly results into better machine learning. It's always important to you prepare the data set properly and during preparation one of the very key step to remember is feature engineering which is nothing but it's a process by which you decide what features you will be passing to the machine learning model so that it can predict the label. So selecting the most suitable features out of available n number of features is also part of this step which we call as preparing the data set. Next step is to select the learning type once you have an idea about the data and the type of features and labels you are predicting you can take a good guess about what type of machine learning would be better for the problem which you are tackling. So here when we say selecting machine learning type it usually indicates whether the person would be going for supervised machine learning or unsupervised or reinforcement learning. And once you select a machine learning type very important is to select and write algorithm. When we are talking about supervised unsupervised there are a lot of algorithms which fall under that category. So based on the data and the features you have and the learning type some algorithms might be very much suitable for the problem. So choosing a right algorithm is one of the very crucial step of machine learning life cycle and generally during this stage most people try to experiment on multiple algorithms they repeat the experiment and then they decide one algorithm out of the algorithms which have been trained on data set. Going ahead once you select an algorithm very important to train your algorithm on training data set and after training we do an evaluation in short what we are talking about we are talking about validation wherein we train on a part of data and we provided data through which your machine learning model did go and we compare the accuracy. So based on accuracy you might go to the earlier step you might change the decision you might choose another algorithm you might change the learning type. So this all forms a small sub cycle in the machine learning life cycle. Going ahead once you have a proper evaluation you would be having a trained ML model you will be deploying it in a production and upon deploying it in a production what you do is you start taking a predictions on a real time data. So this real time data is the one through which your machine learning life cycle might have not gone through earlier. So an actual test of machine learning model which you built is now during the prediction stage and upon the prediction stage very important to make sure you do a proper assessment of prediction take a stock of how your ML algorithm is performing if the accuracy drops down if the accuracy remains constant that could lead to the further decision of repeating the cycle your machine learning life cycle from step one. So this is the machine learning life cycle. So I have a quick question for you. The question is when I talk about testing data set is testing data set unlabeled data set. So you can pause the video here you can revisit the earlier slides which I have discussed in video and you can try to guess the answer for this. But the answer for this question is the testing data is always unlabeled data set. So and the answer is true for it. So this is the bibliography through which I have gone for this video session. And so that's it for this video. Thank you everyone.