WEBVTT 00:00.120 --> 00:02.160 Hello developers, and welcome back. 00:02.160 --> 00:05.240 In this video we're going to learn about machine learning terminology. 00:05.280 --> 00:10.320 We're going to learn about the most used terms in ML. 00:10.400 --> 00:18.880 Let's start with the model in simple terms, the brain or the recipe that the machine learns. 00:19.000 --> 00:24.200 A model is the primary output of the machine learning process. 00:24.520 --> 00:33.400 It's a mathematical function or a program that has been trained to recognize patterns in data. 00:33.640 --> 00:41.520 Once trained, you use the model to make predictions or decisions on new, unseen data. 00:41.680 --> 00:45.360 Think of teaching a child to recognize a dog. 00:45.400 --> 00:48.920 You show them many pictures, the training data. 00:48.960 --> 00:59.640 Their brain forms a pattern of what darkness looks like pointy ears, furry four legs, etc. the learned 00:59.640 --> 01:03.240 pattern in their brain is the model. 01:03.440 --> 01:09.070 When you later show them a picture of a dog they've never seen. 01:09.310 --> 01:14.910 They use their mental model to correctly identify it as a dog. 01:14.950 --> 01:21.830 In technical terms, a model can be a linear regression equation, a decision tree, a neural network, 01:21.830 --> 01:26.950 or any other algorithm that has been fitted to data. 01:26.990 --> 01:29.830 The second thing is feature. 01:29.950 --> 01:37.310 In simple terms, an individual measurable property or characteristic of the data you're analyzing. 01:37.550 --> 01:43.910 Features are the input a variable used by the model to make a prediction. 01:43.950 --> 01:48.750 They are the columns in your data set, excluding the target. 01:48.790 --> 01:55.510 Good features are descriptive and relevant to the problem you are trying to solve. 01:55.710 --> 02:02.390 The process of selecting and creating good features is called feature engineering and is crucial for 02:02.430 --> 02:03.830 model performance. 02:03.830 --> 02:10.150 And this is for advanced topics and more and more dipping deep diving into machine learning. 02:10.270 --> 02:15.150 For example, let's predict house prices, which is the target? 02:15.350 --> 02:20.270 What information you would use to guess the price? 02:20.270 --> 02:22.750 We need to consider square footage. 02:22.790 --> 02:29.190 Number of bedrooms, number of bathrooms, zip code or the location year built. 02:29.270 --> 02:33.590 Each of these pieces of information is a feature. 02:33.630 --> 02:41.710 Another example email spam filter is the email spam or not spam feature the presence of certain words 02:41.710 --> 02:44.670 like free winner, you are a winner you are. 02:44.950 --> 02:51.950 Get your free coupon, the sender's email address, and the number of exclamation marks, etc.. 02:51.990 --> 02:54.230 Okay, so those are the features. 02:54.270 --> 02:56.630 Another terminology is the target. 02:56.630 --> 03:03.310 In simple terms, the answer or the output you want your model to predict. 03:03.310 --> 03:07.790 The target is the variable you are trying to predict or explain. 03:07.990 --> 03:11.550 It's the outcome that depends on the features. 03:11.750 --> 03:19.660 In a data set, it's often a specific column that you ask your model to learn to predict based on all 03:19.700 --> 03:23.780 the other columns in your data set of houses. 03:23.780 --> 03:31.140 Let's take the house prices again in your in your data set of houses, you have columns, for example 03:31.140 --> 03:34.980 square footage, a number of bedrooms, number of bathrooms. 03:34.980 --> 03:36.300 Those are features. 03:36.580 --> 03:39.700 The column sale price is the target. 03:39.700 --> 03:44.500 So we need to get the price of this building or this this house. 03:44.500 --> 03:48.380 So the target is the price and the sale price. 03:48.500 --> 03:55.740 The model's goal is to learn the relationship between the features and this target price. 03:55.740 --> 03:58.940 Again guys, this is the most important thing. 03:59.140 --> 04:01.340 And the link between target and feature. 04:01.380 --> 04:10.500 The model's goal is to learn the relationship between the features and this target price. 04:10.540 --> 04:16.900 Again guys, here we are connecting the model feature and target the models. 04:17.100 --> 04:27.930 The model's goal is to learn the relationship between the features and the target price. 04:27.930 --> 04:31.650 So here is the relationship. 04:31.850 --> 04:33.130 The model's goal. 04:33.170 --> 04:41.330 The idea behind creating the model is to learn the relationship between the feature and the target. 04:41.370 --> 04:45.330 The fourth terminology and term is training. 04:45.530 --> 04:49.890 In simple terms, the process of teaching the model by showing it. 04:49.930 --> 04:50.890 Examples. 04:51.090 --> 04:54.490 Training is the core activity of machine learning. 04:54.650 --> 05:01.090 It's the process where the model learns the relationship between the features and the target. 05:01.210 --> 05:06.050 This is done by feeding the algorithm a large amount of data. 05:06.050 --> 05:13.570 The training data, which includes both the features and the corrected and the correct answers, which 05:13.570 --> 05:14.650 is the target. 05:14.890 --> 05:24.810 The algorithm makes guesses, sees how wrong it is, and then automatically adjusts its internal parameters 05:24.810 --> 05:26.450 to reduce its error. 05:26.610 --> 05:35.610 Training is the process of showing the child hundreds of pictures of dogs and pictures of not dogs, 05:35.610 --> 05:41.490 like cats and rabbits, and telling them which telling them which is which. 05:41.650 --> 05:46.930 Each time they make a mistake, they correct their mental model. 05:47.130 --> 05:52.450 After many examples, their model becomes accurate and well trained. 05:52.450 --> 05:58.530 For a simple model, like a linear regression, training involves finding the line of best fit through 05:58.530 --> 05:59.730 the data points. 05:59.930 --> 06:05.810 The computer tries many different lines and select the one that minimizes the overall distance, which 06:05.810 --> 06:09.370 is the error from all the known data points. 06:09.530 --> 06:20.170 Okay, so as a quick summary feature, the input factors used for prediction target the output we want 06:20.170 --> 06:21.250 to predict. 06:21.290 --> 06:29.130 Training the process of learning the connection between features and target and the model. 06:29.290 --> 06:34.770 The final trained program that makes predictions.