WEBVTT 00:00.480 --> 00:00.780 Hi. 00:01.170 --> 00:09.900 So now we will have a look at your depression project solution, so this particular solution I will 00:09.900 --> 00:19.800 be implementing in CoLab because the data set is pretty large and then we are using Google Kulab. 00:19.800 --> 00:25.830 It will help us run this code a little faster and there is a little chance of crashing and all of the 00:25.830 --> 00:26.850 problems that you face. 00:27.360 --> 00:29.760 So we will be using Google Kulab for this. 00:31.050 --> 00:35.760 Now, a few tips for starting with Google collabs. 00:36.060 --> 00:42.690 You will basically have to open your Google Drive and inside the wheel drive itself. 00:42.690 --> 00:46.110 What you can do is let us say we have this Google Drive. 00:46.380 --> 00:53.220 I can simply say more and inside more. 00:53.220 --> 00:55.370 That is option for Google collaboratory. 00:55.650 --> 01:00.810 So when I open collaboratory, it will be showing me a blank sheet. 01:01.140 --> 01:05.410 And when I see this blank sheet, I can start walking on barbarically. 01:06.120 --> 01:08.280 So here I have this blank sheet. 01:08.280 --> 01:12.030 If I do this, there will be more line of cords. 01:12.030 --> 01:16.890 If I do this, there will be more markdowns which will be added. 01:17.310 --> 01:21.250 Then here we have option for connecting, connecting with the colors. 01:21.300 --> 01:23.760 When we click on this, there are two options. 01:24.090 --> 01:26.490 One is Connector Holsted runtime. 01:26.490 --> 01:37.290 On the other one is going to look one thing so you can use local runtime when you are not really working 01:37.620 --> 01:38.700 on something big. 01:38.940 --> 01:44.700 But if you're working on a larger data set or when you want to have faster processing, you can use 01:44.790 --> 01:45.710 those different thing. 01:46.110 --> 01:48.480 And here you have options for monetization. 01:48.490 --> 01:53.970 So when I click on this, it shows me different kind of sessions which are open and then you can decide 01:53.970 --> 01:56.490 if you want to close a session or something like that. 01:57.810 --> 02:00.510 Next thing which we have here is, again, in the done thing. 02:00.900 --> 02:07.020 We have this a change runtime option, which allows me to have these three options. 02:07.020 --> 02:10.050 One is none under the one you wanted of one is to use. 02:10.770 --> 02:14.850 Nine is when you are not using any hardware accelerator. 02:14.850 --> 02:17.820 That is, you don't want to foster anything. 02:17.820 --> 02:23.040 You want to use it just like you use your normal laptop or your computers. 02:23.400 --> 02:26.610 Then you want a little faster processing. 02:26.610 --> 02:31.230 You can use abuse and abuse goes on. 02:31.230 --> 02:34.470 The abuse do not have much difference. 02:34.890 --> 02:42.110 But when you are working on deep learning or when you're working with dancers at that point of time, 02:42.120 --> 02:47.010 the abuse would perform better because they are working on dancers. 02:47.020 --> 02:51.840 And so that is when you can use Depuis otherwise working with your workspace. 02:53.820 --> 02:59.100 So I will be using none for now and I click save and then it will connect for me. 02:59.460 --> 03:01.320 I just click connect and it will connect. 03:01.530 --> 03:03.570 So now going back to our solution 03:07.110 --> 03:09.920 now here we have this code. 03:09.930 --> 03:17.840 So I the first thing that you will be doing is running this particular command which will mount your 03:17.850 --> 03:18.560 whole bedrick. 03:18.960 --> 03:22.410 So basically when I run this, what happens is I did run this for you. 03:35.380 --> 03:39.050 It will ask me for this particular warrant. 03:39.070 --> 03:45.490 So when I click this particular, you are eligible, ask me to select one particular email, idy or 03:45.490 --> 03:48.670 one particular drive from which I want to retrieve the data. 03:49.900 --> 03:56.470 So once I choose my email I.D., it will give me this particular option where I can see a lull and after 03:56.470 --> 04:01.270 I see allow it will give me this particular code, which I can simply copy by clicking on this. 04:01.690 --> 04:04.180 And I will go back and then I go back. 04:04.180 --> 04:07.690 I enter this phase out of here and then I enter this. 04:07.690 --> 04:11.480 It will authorize this and connect with the Google Drive. 04:12.400 --> 04:19.390 So once this will drive this connected, if you click on this Fi's option, you can see the entire drive 04:19.390 --> 04:19.930 structure. 04:19.960 --> 04:25.060 So if I click on this, I be able to see all the different files which are present in my drive and all 04:25.060 --> 04:25.780 those details. 04:28.120 --> 04:30.700 So this is what I will be having. 04:30.730 --> 04:39.340 So what I can do is I can simply go to this particular run by different libraries, which I want to 04:39.340 --> 04:39.970 import. 04:40.420 --> 04:44.380 Then I can provide the CSP details. 04:44.560 --> 04:48.260 So I have to provide the parts for the CSP file. 04:48.520 --> 04:51.660 So let me show you how I got this particular box. 04:53.860 --> 05:01.390 So I just went inside this particular part in the files and I think here and then I'll write click and 05:01.390 --> 05:04.240 I copy both and it will give me the part for this. 05:04.810 --> 05:07.260 That's all you need to do to import any fight. 05:08.170 --> 05:09.160 After that. 05:09.160 --> 05:12.100 I have imported this data set. 05:12.110 --> 05:20.530 So this data set contains different ideas, the hazard details, which is the numeric value, which 05:20.530 --> 05:30.170 isn't in Dederer value, which is a whole number, which is not a floating point number, basically. 05:30.910 --> 05:34.960 Then we have different variables which we have to keep off the book. 05:35.740 --> 05:40.480 Now out of these variables, if I see the shape there, around 40000 rows of data. 05:40.480 --> 05:46.180 Thirty four columns, I will find out the target column as Hassard. 05:46.420 --> 05:49.750 And this idea does not make really sense to me. 05:49.750 --> 05:53.980 It doesn't give much information so I can really get rid of this column. 05:55.240 --> 05:58.690 Next is the ANOVA one we do with the 50. 05:58.990 --> 06:04.080 I will have to take care of these columns and find out which are actually important and which are not. 06:04.390 --> 06:07.290 So I will get the column five. 06:07.300 --> 06:13.660 So if we see the airport select detail and find out the object type, you can see that these all are 06:14.650 --> 06:20.850 categorized as Object B, so I will find out the cutoff value. 06:20.860 --> 06:24.210 So here I am sitting because of value to be five percent. 06:24.730 --> 06:32.860 So whatever value is having five percent, less than five percent frequency of any information internally, 06:33.220 --> 06:37.570 I will simply get rid of those particular categories. 06:37.570 --> 06:42.550 I convert my category categorical columns into dummy variables. 06:44.020 --> 06:47.740 So this is the code which we have been using for ever. 06:48.070 --> 06:55.240 So it's simply the ject me, the categories which have frequency more than the cutoff value, which 06:55.240 --> 06:58.140 is more than five percent in any particular column. 06:58.510 --> 07:08.110 And then I get these different categories and it converts these categorical variables into dummy columns. 07:08.180 --> 07:10.980 So these are the columns which have been generated. 07:11.350 --> 07:18.460 Now next thing which we will be doing is I have just included define this particular report function, 07:18.460 --> 07:20.830 which will give me the report next. 07:20.840 --> 07:27.550 What I do is I create this VYE data frame, which contains my target and state of name, which contains 07:27.550 --> 07:36.910 my features or attributes next time reporting different libraries for my model building like Bensedrine 07:36.910 --> 07:37.990 split randomise. 07:38.000 --> 07:45.640 So it's the only decision created this linear regression, which all is the next thing which I do is 07:45.640 --> 07:51.880 I am simply implementing the ridge regression here. 07:51.880 --> 07:59.920 Ridge regression is giving me all these different runs and later, towards the end of. 08:03.790 --> 08:09.550 I can see this information which has been retrieved, and the best estimate that comes out to be the 08:09.550 --> 08:12.400 one with Elfy is equal to one hundred makes. 08:12.400 --> 08:19.290 What I do is I can find out the mean validation school, which is minus two point seven seven five. 08:19.600 --> 08:27.700 Now, at the beginning of this particular project, we had a guy claim that the scoring will be done 08:27.880 --> 08:36.220 based on a particular formula, which you must have noted down, which is one minus the mean validation 08:36.230 --> 08:36.850 score. 08:36.850 --> 08:43.320 That is the error of divided by a five point four. 08:43.330 --> 08:44.890 So that is what we are looking for. 08:44.890 --> 08:49.590 We are looking for a value greater than five point one there. 08:50.050 --> 08:55.240 So these are the different results, which are not really satisfactory enough for us. 08:55.660 --> 09:05.230 So we will go ahead and we will load the predictions versus the data we want, which is one of the variables 09:05.230 --> 09:06.170 which are present here. 09:06.190 --> 09:14.040 So if you see the believin is scattered like these, I'm the prediction, which we have our prison here. 09:14.710 --> 09:22.970 So you can see immediately it is not able to find those these values which are scattered about. 09:24.400 --> 09:32.510 Similarly, that we will go further and implement the decision created this year and inside this decision 09:32.570 --> 09:39.790 to said again, we are finding out they may not be negative, mean, absolute error, and we have implemented 09:39.790 --> 09:41.080 this entire thing. 09:41.080 --> 09:46.690 And again, this is the main validation score which we obtained out of it. 09:47.530 --> 09:49.770 And these are the predictions. 09:49.790 --> 09:53.000 So now you can see this is already better than the previous one. 09:54.100 --> 09:58.290 The next thing which we do is we are implementing the random for this trip this year. 09:58.870 --> 10:06.690 So as we go further, it does find something relevant here. 10:24.740 --> 10:32.750 So here we are using the land of what is vessel and we are using negative, mean absolute data and again, 10:33.170 --> 10:40.790 it has this particular desert, which it gives us out of it, and they show the desert to. 10:43.640 --> 10:45.130 This is the Red Crescent. 10:45.160 --> 10:53.660 This is the other details, and now here we are implementing the most aggressive with the objective, 10:53.870 --> 10:55.100 as we saw. 10:55.410 --> 11:02.630 Now, this is what will actually help us achieve our target, because earlier what we were doing was 11:02.630 --> 11:06.320 while we were targeting, we were finding out cantinas values. 11:06.320 --> 11:14.290 But when we use conquest, only me give us the values which are not in a floating point of format. 11:14.600 --> 11:15.910 So we'll go further. 11:15.920 --> 11:19.730 And here you can see that the result is minus two point seven to six. 11:20.300 --> 11:27.020 And as we go further, you can see the results have further improved. 11:27.020 --> 11:29.780 Move now. 11:29.840 --> 11:34.330 Next, we will go on doing the barometer here. 11:34.340 --> 11:42.340 If you would have noticed one thing they are doing the sequentially cleaning, which is we asked for 11:42.350 --> 11:49.510 in the first run, the fix the estimate when the object, all of the barometers are commented here, 11:50.120 --> 11:52.060 very ladies are dad. 11:52.070 --> 11:56.930 It will consider these values to be the default values for the body which are already present. 11:57.590 --> 12:02.790 So when we wrap this, we are trying to find out the estimated value when hit on this. 12:02.790 --> 12:06.340 Here I find out the estimated value 500 folks built for me. 12:06.890 --> 12:08.000 So I go ahead. 12:09.320 --> 12:19.730 And in the next one I put in estimate Demetri's 250 and further, I try to find out different learning 12:19.760 --> 12:22.820 rates for myself now, different learning rate. 12:22.830 --> 12:28.220 So I have provided all these learning rates and I'm running this training, this model again. 12:28.940 --> 12:40.340 When I put this model and I see the results, it allows me to find out of more better values. 12:40.400 --> 12:48.760 And I'm able to find out that learning rate, which is doing well for me, is zero point zero zero five. 12:49.610 --> 12:51.370 So I will fix this again. 12:51.890 --> 12:57.470 Now, I fixed my loan exactly zero point zero zero five and the next best one, that's zero point zero 12:57.470 --> 12:59.610 zero two to there. 12:59.620 --> 13:01.250 Also, I have all fixed. 13:01.700 --> 13:06.990 Now, what I will do is I am training this to find out the best muxtape. 13:07.760 --> 13:13.000 So now again, I will fit the mold of the model here. 13:13.550 --> 13:19.370 We get to know that this is the desired for us. 13:21.170 --> 13:28.610 So we go further and you can see that the learning tool is performing well in maximum depth. 13:28.760 --> 13:29.830 Seven performance. 13:30.620 --> 13:31.550 So we go ahead. 13:32.990 --> 13:38.590 We put the maximum that the seven learning theta zero point zero zero two and estimated about 500. 13:38.600 --> 13:43.210 These are the values which we have fixed and subsample is the values which we are trying to find out 13:43.220 --> 13:43.410 now. 13:43.640 --> 13:50.480 Again, we will do the same thing and I predict we will improve our model and fix the parameters. 13:50.910 --> 13:57.240 Now, doing this is useful only if your model performance also improves. 13:57.980 --> 14:03.520 So if my modeling performance is not improving, so this value is minus two point sixty six. 14:04.490 --> 14:15.080 If I go for the standard deviation of zero point three four nine five, if I go above here, we have 14:15.560 --> 14:18.000 minus two point six to seven. 14:18.590 --> 14:21.080 So here you can see there is a slight improvement. 14:21.560 --> 14:27.950 But I will keep on going only if this there is a good number, good amount of improvement. 14:28.190 --> 14:30.760 It is improving by zero point one percent. 14:31.580 --> 14:33.550 So that is not much improvement. 14:33.560 --> 14:33.830 Right. 14:34.070 --> 14:37.010 So we are not looking for that small amount of improvement. 14:37.010 --> 14:38.480 We are looking for a large and improvement. 14:38.510 --> 14:41.630 So anybody from this would be fine for us. 14:42.860 --> 14:47.300 Now, what you can do is you can create different models. 14:47.300 --> 14:51.050 So this is just one of the examples I used. 14:51.300 --> 14:58.190 You can start different models and you use count for something that I don't see how you can improve 14:58.190 --> 15:05.420 that you can create a combination of different models and try implementing those.