1 00:00:00,780 --> 00:00:07,960 ‫In the last lecture we saw how to create how to combine and how to screen our classification more than 2 00:00:08,260 --> 00:00:12,730 ‫intensive look at us in this video. 3 00:00:12,820 --> 00:00:19,500 ‫We will be and train a regression model using gave us for this. 4 00:00:19,520 --> 00:00:29,380 ‫We will be using the very popular regression dataset that is California housing data set this dataset 5 00:00:29,830 --> 00:00:35,100 ‫is available in a Skillern database library. 6 00:00:35,110 --> 00:00:42,460 ‫The objective here is to predict the prices of homes using a different independent variables. 7 00:00:43,540 --> 00:00:45,690 ‫So let's get started. 8 00:00:45,820 --> 00:00:56,210 ‫First we are importing some basic liabilities such as Num by on does and my lib then we are importing 9 00:00:56,220 --> 00:01:06,860 ‫tens of flowing gave us and then since this data is available in and school on dataset we are also importing 10 00:01:06,860 --> 00:01:14,720 ‫California housing from escalating dataset and we are saving this database and going another way label 11 00:01:15,320 --> 00:01:18,090 ‫housing. 12 00:01:18,090 --> 00:01:23,940 ‫I also want to share one small shortcut to access the help of any function. 13 00:01:23,940 --> 00:01:31,980 ‫So if you just click in between of any function parenthesis and then hold leadership key and click on 14 00:01:31,980 --> 00:01:32,580 ‫tab. 15 00:01:32,700 --> 00:01:40,910 ‫So if you had ship plus that it will open the help or documentation of that function. 16 00:01:41,280 --> 00:01:49,170 ‫If you shift type food banks it will expand their documentation you can see here. 17 00:01:49,340 --> 00:01:54,830 ‫This function will return different parameters such as not data. 18 00:01:54,830 --> 00:02:02,520 ‫It will give us the independent variables that target will give us the dependent variable and the feature 19 00:02:02,560 --> 00:02:09,590 ‫me will give us the details of features let's just close this. 20 00:02:09,590 --> 00:02:16,500 ‫So if you have any doubt regarding any function just click between parenthesis and head shift. 21 00:02:16,550 --> 00:02:25,920 ‫Let's add so I have alluded listed some of the information about this database. 22 00:02:25,920 --> 00:02:30,620 ‫So in this database there are around don't polls and records. 23 00:02:30,780 --> 00:02:38,270 ‫There are eight independent variables we have first variable as mad in. 24 00:02:38,720 --> 00:02:43,160 ‫This is the medium income in that particular block warehouses located. 25 00:02:44,090 --> 00:02:47,410 ‫Then we have a second variable that is House 8. 26 00:02:47,450 --> 00:02:51,550 ‫This is the median house each in that block. 27 00:02:51,560 --> 00:02:59,510 ‫Then we have average rooms which is the average number of rooms and then we have average bedrooms for 28 00:02:59,510 --> 00:03:01,670 ‫average number of bedrooms. 29 00:03:01,700 --> 00:03:07,670 ‫Next we have a population variable that is the block population. 30 00:03:07,670 --> 00:03:10,490 ‫Then we have average occupancy. 31 00:03:10,490 --> 00:03:19,590 ‫That is the average house occupancy and then latitude and longitude of that holds block using this eight 32 00:03:19,620 --> 00:03:21,480 ‫independent variables. 33 00:03:21,480 --> 00:03:24,970 ‫We want to predict the value of the house. 34 00:03:25,020 --> 00:03:28,060 ‫The values are in hundred thousands. 35 00:03:28,110 --> 00:03:30,710 ‫So supposing forward why variable is five. 36 00:03:30,720 --> 00:03:39,390 ‫That means the value of that houses five hundred thousand dollars so this is our dataset. 37 00:03:39,440 --> 00:03:45,190 ‫We have this independent variable and one target variable as base. 38 00:03:45,650 --> 00:03:52,340 ‫If you want some more detail about this dataset you can click on this documentation link. 39 00:03:52,700 --> 00:03:57,940 ‫This will open the official Eskil on documentation of this dataset here. 40 00:03:58,100 --> 00:04:06,360 ‫You will get to know about all the parameters that you can give and what are included in this database. 41 00:04:06,520 --> 00:04:07,970 ‫Let the school back 42 00:04:11,000 --> 00:04:20,970 ‫so this housing is in the form of a dictionary and we have one devalued beer as feature name. 43 00:04:21,860 --> 00:04:25,950 ‫So just look at the feature names first. 44 00:04:26,690 --> 00:04:35,280 ‫You can see these are the variable names that we have discussed already now to access the independent 45 00:04:35,280 --> 00:04:35,870 ‫data. 46 00:04:36,060 --> 00:04:42,200 ‫We have to use housing data and to access the independent data set. 47 00:04:42,330 --> 00:04:44,450 ‫We have to use housing data gate 48 00:04:47,430 --> 00:04:58,110 ‫so in this line of code we are splitting our data first in blue green full and best data set then we 49 00:04:58,110 --> 00:05:07,540 ‫add further dividing this green full dataset and do X green and x y new addition dataset. 50 00:05:08,210 --> 00:05:16,850 ‫We will be using test Green is from Eskil on model selection and then we will use best green display 51 00:05:17,090 --> 00:05:20,840 ‫metric to divide our data. 52 00:05:20,840 --> 00:05:25,320 ‫We are not giving any additional parameter for test sites. 53 00:05:25,310 --> 00:05:32,110 ‫That's because by default the test size is 25 percent of the total data. 54 00:05:32,210 --> 00:05:42,590 ‫So 25 percent of the total data that is around 20000 goals will go into test set and then the remaining 55 00:05:42,590 --> 00:05:52,430 ‫75 percent will go in training site all of that 75 percent again 25 percent will go into validation 56 00:05:52,430 --> 00:05:57,010 ‫set and rest of 75 percent will go into our green site. 57 00:05:58,280 --> 00:05:59,780 ‫Let's run this 58 00:06:02,690 --> 00:06:11,450 ‫next step is to process our data and we will be using the standard is scalable from a scalar to standardize 59 00:06:11,450 --> 00:06:15,670 ‫our data and standardizing. 60 00:06:15,980 --> 00:06:24,080 ‫We subtract the mean of each variable from their individual values and then we also divide it by the 61 00:06:24,080 --> 00:06:34,280 ‫variance because at the end we want all the variables with mean zero and their variance as one. 62 00:06:34,340 --> 00:06:42,520 ‫This is a standard procedure to create and the machine learning more than the steps here are very simple. 63 00:06:42,560 --> 00:06:51,110 ‫First we are importing standard skill it from a skill and processing then we are creating a scalar object 64 00:06:51,920 --> 00:07:00,170 ‫using standard scalar method then we are creating this scalar object using what extreme data. 65 00:07:01,910 --> 00:07:09,440 ‫So on our extended data this is scalar will find the values to subtract as I mean and to divide as a 66 00:07:09,440 --> 00:07:10,330 ‫variance. 67 00:07:10,640 --> 00:07:17,870 ‫Then we will use those values or this is scalar object to standardize our validation and as that as 68 00:07:17,870 --> 00:07:22,280 ‫well just so repeat we are fitting. 69 00:07:22,280 --> 00:07:28,540 ‫This is scalar object on our training data and we are transforming our validation and test set. 70 00:07:28,550 --> 00:07:36,190 ‫Using this is scalar object that we have to take on our extreme data look for it. 71 00:07:37,110 --> 00:07:44,780 ‫We will be using it underscore transform method and we will be using extreme as parameter to transform. 72 00:07:44,780 --> 00:07:52,210 ‫We will be using dot transform method of this scalar object and we will be using relevant datasets here 73 00:07:55,740 --> 00:08:00,330 ‫so let's create a standardised datasets. 74 00:08:00,410 --> 00:08:03,700 ‫We are saving this object in their original name only. 75 00:08:03,780 --> 00:08:08,970 ‫So we are replacing the ordinal extreme with this standardised version of extreme. 76 00:08:08,970 --> 00:08:15,330 ‫Our order is null ex validation set in a standardised version of X in the relation set and same for 77 00:08:15,380 --> 00:08:17,160 ‫this site as well. 78 00:08:17,420 --> 00:08:24,760 ‫It says this If you want some more information among this the Skilling you can always do to as skill 79 00:08:24,790 --> 00:08:29,230 ‫and documentation center disclaimers. 80 00:08:29,370 --> 00:08:33,700 ‫Next step is to set random seeds. 81 00:08:33,840 --> 00:08:37,740 ‫This is to generate the same result every time we run this modern 82 00:08:42,220 --> 00:08:42,800 ‫now. 83 00:08:42,820 --> 00:08:50,410 ‫As I said earlier our initial dataset was off or on point deposing rules or records. 84 00:08:50,410 --> 00:08:57,530 ‫Now let us see the shape of the extended asset. 85 00:08:57,560 --> 00:09:06,980 ‫Here you can see we have eight columns and don't let thousand six hundred records in our extremely desolate 86 00:09:08,950 --> 00:09:20,330 ‫we should never own five puzzling records in our X dataset and around 4000 in valuation say. 87 00:09:20,780 --> 00:09:29,360 ‫Now let's create the structure for our regression neural network here. 88 00:09:29,390 --> 00:09:37,970 ‫We will be first having an input layer then we will be having the first dense layer with neurons. 89 00:09:38,180 --> 00:09:43,410 ‫Then we want to create second dense layer with another 30 neurons. 90 00:09:43,760 --> 00:09:52,220 ‫And then since this is a regression problem we will be having a single output neuron without any activation 91 00:09:52,220 --> 00:09:54,060 ‫function. 92 00:09:54,620 --> 00:10:03,520 ‫A single neuron since we want a continuous value as our output again we will be using the sequential 93 00:10:03,610 --> 00:10:17,950 ‫EPA we are saving this structure or our model as model and then for the first layer we are writing gave 94 00:10:17,960 --> 00:10:22,430 ‫us dot layer dot dense here in parenthesis. 95 00:10:22,430 --> 00:10:26,270 ‫We have to provide the number of neurons which is 30. 96 00:10:26,320 --> 00:10:32,080 ‫Then as discussed in our two re lectures we will be using activation function as the loop. 97 00:10:32,450 --> 00:10:40,280 ‫And then since this is our first tool on there we need to provide the input shape since the number of 98 00:10:40,400 --> 00:10:44,230 ‫independent variables in our data is eight. 99 00:10:44,290 --> 00:10:51,860 ‫We will be using good shape equal to. 100 00:10:52,130 --> 00:11:00,260 ‫You can also use like this input shape will too extreme dot shape and then calling the second and so 101 00:11:00,260 --> 00:11:03,820 ‫on elements of over input chip. 102 00:11:03,860 --> 00:11:09,550 ‫This way you don't have to worry about changing this number every time you change your data base. 103 00:11:09,650 --> 00:11:17,870 ‫You can just trade extreme not shape and it will automatically get the number of variables from the 104 00:11:17,870 --> 00:11:23,580 ‫shape attribute of our extreme object. 105 00:11:23,680 --> 00:11:29,170 ‫So this is the structure of our first dance layer will be cleared. 106 00:11:29,170 --> 00:11:31,570 ‫Second dense layer in a similar fashion. 107 00:11:31,780 --> 00:11:39,610 ‫We will be using US DOT layer not dance and then the number of neurons in the parenthesis which is 30 108 00:11:40,030 --> 00:11:46,270 ‫and activation function as a low similarly for the output layer. 109 00:11:46,410 --> 00:11:50,080 ‫We will be using data as Dot players or dance. 110 00:11:50,130 --> 00:11:56,550 ‫And since this the regression problem we will be using the single neuron without any activation function 111 00:11:58,670 --> 00:12:00,870 ‫just run this. 112 00:12:01,580 --> 00:12:09,590 ‫And again one important thing you can come in using hash symbol inside the cells. 113 00:12:09,590 --> 00:12:17,050 ‫So python will execute on this part of the code and it will not be executing any code which starts with 114 00:12:17,060 --> 00:12:21,630 ‫hash as the loop then for starting the coming. 115 00:12:21,700 --> 00:12:30,280 ‫So now we have created the structure or architecture of the word neural network on just look on form 116 00:12:30,280 --> 00:12:31,640 ‫and view this structure. 117 00:12:31,680 --> 00:12:35,800 ‫We can call thought somebody might have moderated or 118 00:12:40,890 --> 00:12:47,490 ‫somebody here you will get the information about the structure that we have created. 119 00:12:47,670 --> 00:12:51,050 ‫So we have first dense layer with 13 neuron. 120 00:12:51,320 --> 00:12:53,890 ‫We have second dense layer with pretty neurons. 121 00:12:53,940 --> 00:13:00,450 ‫And lastly we have a single output layer with one neuron. 122 00:13:00,480 --> 00:13:02,520 ‫This is what we want. 123 00:13:03,180 --> 00:13:09,750 ‫The next step should be to compile this Martin again. 124 00:13:09,790 --> 00:13:15,870 ‫The combined method works similar for both classification and duration model. 125 00:13:16,000 --> 00:13:19,930 ‫First we have to mention the loss in classification. 126 00:13:19,930 --> 00:13:27,490 ‫We were using cross entropy but here since we are running regression we have to use means squared added 127 00:13:28,540 --> 00:13:36,380 ‫also known as MVC the second parameter is optimizer again. 128 00:13:36,580 --> 00:13:41,960 ‫Here also we are using as Zuhdi spastic gradient descent. 129 00:13:42,370 --> 00:13:50,260 ‫And here we are also providing the learning rate by default the value of learning rated zero point zero 130 00:13:50,260 --> 00:13:53,670 ‫one and two change. 131 00:13:53,740 --> 00:13:57,990 ‫You can just write the new value in the parent. 132 00:13:58,040 --> 00:14:05,070 ‫This we have already discussed what is learning rate in auditory lecture. 133 00:14:05,080 --> 00:14:08,350 ‫So if you have any doubts just read the entire lecture. 134 00:14:09,930 --> 00:14:11,080 ‫And then the next. 135 00:14:11,070 --> 00:14:13,590 ‫But I think that we are passing these metrics. 136 00:14:13,590 --> 00:14:18,680 ‫This is an optional parameter in classification. 137 00:14:18,680 --> 00:14:25,190 ‫We are using accuracy but in regression we can use mean absolute. 138 00:14:25,200 --> 00:14:35,540 ‫Edit R M E absolute error is the difference between the predicted value and the actual value whereas 139 00:14:35,720 --> 00:14:36,770 ‫the mean is squared. 140 00:14:36,770 --> 00:14:40,860 ‫Edit is the square of that difference. 141 00:14:41,390 --> 00:14:49,160 ‫So we are calculating both mean squared error as a lost function and add me as the metrics we additionally 142 00:14:49,160 --> 00:14:50,030 ‫want to calculate. 143 00:14:51,490 --> 00:15:00,250 ‫Again just remember if you want to look at the documentation or hell just click inside of the parenthesis 144 00:15:00,580 --> 00:15:07,910 ‫and and press shift less tab you will get this kind of documentation. 145 00:15:07,920 --> 00:15:13,700 ‫And here you can see that by default the learning rate value is 0.01. 146 00:15:14,000 --> 00:15:21,230 ‫So in classification we did not provided any learning rate so the learning rate that was used there 147 00:15:21,410 --> 00:15:23,220 ‫was 0.01. 148 00:15:24,260 --> 00:15:28,940 ‫But you can always change these values into your named. 149 00:15:29,300 --> 00:15:39,430 ‫Let us run this so we have compiled our more than the next step the next step is to print out more. 150 00:15:39,690 --> 00:15:44,760 ‫Using training data the method or the process is same. 151 00:15:44,940 --> 00:15:49,480 ‫We are creating and another object model underscoring street for creating. 152 00:15:49,530 --> 00:15:57,930 ‫Then we are using model not fit and we are passing our training dataset number of epochs and the validation 153 00:15:57,940 --> 00:16:00,640 ‫dataset that we have created. 154 00:16:00,980 --> 00:16:02,940 ‫Just this a statement 155 00:16:05,880 --> 00:16:10,870 ‫again just like the classification model. 156 00:16:11,080 --> 00:16:15,640 ‫You will get the lost value which is the meaning of square data. 157 00:16:15,910 --> 00:16:20,870 ‫You will get the end me when you mean absolute error. 158 00:16:21,340 --> 00:16:26,390 ‫And similarly you will get these two values for your validation set that's when 159 00:16:30,080 --> 00:16:39,520 ‫and you can see that the loss on both training set and validation set is decreasing with each epoch. 160 00:16:39,540 --> 00:16:44,600 ‫Now we have these values for our training and validation set. 161 00:16:44,700 --> 00:16:52,500 ‫We can also evaluate performance of this grain model on our test set and we are going to use the same 162 00:16:52,500 --> 00:16:58,350 ‫method as we did with classification Martin will call model not evaluate. 163 00:16:58,470 --> 00:17:00,630 ‫And then we will pass our training there. 164 00:17:00,650 --> 00:17:10,420 ‫I say this then this you can see on our training does the loss is zero point three. 165 00:17:10,420 --> 00:17:22,760 ‫That is MFC or mini squared at it and m e e mean absolute error is zero point 4 4 9 3 No. 166 00:17:22,770 --> 00:17:30,520 ‫Just in classification we can call model history dot history that will give us the values of all these 167 00:17:30,520 --> 00:17:40,770 ‫my crisis in a formal dictionary and this year you will get the loss and add me on training dataset 168 00:17:41,100 --> 00:17:48,710 ‫and validation loss and validation and me the beauty of this is we can load this dictionary on a block 169 00:17:49,210 --> 00:17:57,300 ‫and just like we did for classification and that will show us how our training loss and regulation loss 170 00:17:57,390 --> 00:18:03,490 ‫are changing with each epoch and whether we have achieved the convergence or not. 171 00:18:04,930 --> 00:18:14,480 ‫Just this so you can see we have the lost values and the Emmy values for over framing and validation 172 00:18:14,480 --> 00:18:23,180 ‫set noted on this graph and one thing to notice is this graph is still going down meaning that if we 173 00:18:23,180 --> 00:18:32,520 ‫run some more epochs this will further decrease the losses and improve the accuracy of our more. 174 00:18:33,230 --> 00:18:39,830 ‫So this is the one way to tell whether you have achieved convergence or not or whether you have been 175 00:18:39,830 --> 00:18:46,410 ‫pleased you are awkwardly or not you have to look at this validation loss and validation Emmy value 176 00:18:47,340 --> 00:18:57,080 ‫so this the validation lost and you can clearly see that it is going down so to improve the accuracy 177 00:18:57,560 --> 00:19:04,200 ‫we can read on the score to run it for finding more epochs 178 00:19:09,140 --> 00:19:19,670 ‫so just go there now one important thing about guitars these guitars have the beats and biases value 179 00:19:19,880 --> 00:19:21,020 ‫in the memory. 180 00:19:21,020 --> 00:19:29,480 ‫So if you just read on this whole segment again this will not screen the data from Isaak but it will 181 00:19:29,570 --> 00:19:34,250 ‫start draining the data from this position. 182 00:19:34,790 --> 00:19:42,090 ‫So if we run this a statement bu times that is similar to running this statement for the epochs. 183 00:19:42,680 --> 00:19:53,470 ‫If we just read on one more time you can see earlier the lost values were around point seven nought 184 00:19:53,480 --> 00:19:58,010 ‫point take for the first sequel and then gradually decreasing. 185 00:19:58,240 --> 00:20:02,100 ‫But now we have started from the 20 plus epoch 186 00:20:07,630 --> 00:20:11,800 ‫last time the lost value on our test set was 0 1 3 0. 187 00:20:11,810 --> 00:20:22,480 ‫Let's see whether we have improved this lost value or not and see the losses decrease from 0 1 3 0 0 188 00:20:22,480 --> 00:20:25,160 ‫point to 5. 189 00:20:25,210 --> 00:20:34,720 ‫So our hypothesis was correct that the model was not converged in 2012 there was room of improvement 190 00:20:34,960 --> 00:20:38,470 ‫and be lead in the whole more than 420 more epochs. 191 00:20:38,470 --> 00:20:40,570 ‫That is a total loss for the box 192 00:20:43,610 --> 00:20:50,820 ‫and we can see this graph in just focus on this regulation. 193 00:20:50,820 --> 00:21:02,800 ‫Lastly earlier it was zero point for all we read and this there is a slight decrease in valuation loss. 194 00:21:03,020 --> 00:21:06,880 ‫Now you can see that the line is flattened turned out. 195 00:21:06,890 --> 00:21:12,440 ‫This means we have achieved the convergence on this model. 196 00:21:12,440 --> 00:21:14,690 ‫So not just with regression. 197 00:21:14,780 --> 00:21:20,930 ‫If you are running classification model as well just look at this graph to identify whether you have 198 00:21:20,990 --> 00:21:22,680 ‫achieved convergence or not. 199 00:21:24,290 --> 00:21:30,820 ‫Now to predict the values on the new dataset you can always use credit market. 200 00:21:31,370 --> 00:21:34,390 ‫So your object name and not really matter. 201 00:21:34,760 --> 00:21:37,170 ‫And then the new dataset. 202 00:21:37,310 --> 00:21:45,110 ‫I don't have any new dataset so I am just creating the sample of 4 3 values of my X asset and considering 203 00:21:45,200 --> 00:21:54,470 ‫it as my new dataset and then saving the information in y predicted values and using model lot predict 204 00:21:54,530 --> 00:21:57,010 ‫my total to predict the values. 205 00:21:57,340 --> 00:22:01,710 ‫These are the values using this model. 206 00:22:01,760 --> 00:22:04,670 ‫That's all for this lecture in the next lecture. 207 00:22:04,670 --> 00:22:09,720 ‫We will be looking at the functional API of get us. 208 00:22:09,830 --> 00:22:10,210 ‫Thank you.