WEBVTT 00:01.340 --> 00:08.130 High in this session, we will discuss why we should consider randomized search instead of research. 00:08.780 --> 00:15.330 So here the variable bottoms below consists of five different barometers we intend to do. 00:16.000 --> 00:23.650 Now, each of these parameters contains some values he had last week, contains none and balanced criterion. 00:23.660 --> 00:25.580 Fancy contains entropy. 00:25.580 --> 00:34.370 Guiney Muxtape has values none and five to seven B means, I believe, has values one to five twenty. 00:34.640 --> 00:39.670 Then we have Montabaur split, which has values to five, 10, 15 and 20. 00:40.100 --> 00:45.660 Now the possible combinations will be nine sixty as shown below. 00:45.920 --> 00:53.780 If we calculate this, if we come out to be nine sixty, which is to cross to cross eight. 00:54.410 --> 00:56.210 Six across five. 00:57.980 --> 01:05.870 Now, if the use grid search and use up, then for TV, we will build around ninety six hundred individual 01:05.870 --> 01:13.940 trees, it will result in the best possible combination but will also take a lot of time in order to 01:13.950 --> 01:14.960 handle this. 01:15.140 --> 01:22.370 Instead of drying out all 960 combinations, we can try only 10 percent of these combinations. 01:22.400 --> 01:24.800 That is 96 combinations. 01:25.250 --> 01:34.130 Randomized search will randomly selected six of the combinations and will result in good enough result, 01:34.820 --> 01:41.780 though we do not have a guarantee of the best combination, but it will give us a good enough combination 01:41.780 --> 01:43.340 at a fraction of the done thing. 01:44.000 --> 01:53.870 The tradeoff between runtime and how good a result will be is this particular randomized search wasis 01:53.870 --> 01:54.540 grid search. 01:55.130 --> 02:02.870 Now we can make randomized search better by running it multiple times each time we get a different combination 02:02.870 --> 02:09.860 and check whether the combinations are consistent across different fronts like this, we can expand 02:09.860 --> 02:12.000 in the neighborhood values as well. 02:12.020 --> 02:16.040 For example, we get maximum depth of 70 using randomized search. 02:17.170 --> 02:26.020 So the maximum value of Muxtape is actually seven feet now in the next run, we would want to add 80 02:26.020 --> 02:27.560 and 90 to check again. 02:28.360 --> 02:35.050 Another example would be if we get maximum depth has 30, but we did not consider any value between 02:35.050 --> 02:36.160 30 and 50. 02:37.350 --> 02:41.970 So we may want to add a maximum depth of 40 and try again. 02:43.050 --> 02:50.880 So like this, once we get the best maximum depth value as Astarte, we can also try values around Kadafi 02:50.880 --> 02:56.730 like twenty five, twenty six, thirty one thirty, a sector which may result in a better performance 02:56.730 --> 02:57.390 on the model. 02:57.840 --> 02:58.890 So this is how. 02:59.960 --> 03:09.850 We can actually use randomizer, TV on the TV now let us see how randomizer TV works so it will look 03:09.860 --> 03:19.070 just like TV, except that we need to tell randomizer TV that out of all of these combinations, how 03:19.070 --> 03:20.850 many do we want to try out? 03:21.560 --> 03:26.330 So here we will provide the value of the night of. 03:27.490 --> 03:36.220 So what we will do is we will mention that out of mind, 60 combinations only we pry open, we use the 03:36.220 --> 03:38.200 argument and who said this? 03:38.530 --> 03:42.750 I ideally they should try about 10 to 20 percent of all the combinations. 03:42.940 --> 03:48.880 So accordingly, we can said disvalue of inflation, and then we get a particular value. 03:49.090 --> 03:56.980 We can just leave on the value and change the value and try different combinations and retry it multiple 03:56.980 --> 04:01.150 times so that we are satisfied with the desired, which we have obtained.