WEBVTT - This file was automatically generated by VIMEO

0
00:00:07.990 --> 00:00:12.330
Hi. In this video we will learn more about, uh, randomized search.

1
00:00:15.600 --> 00:00:20.110
Randomized search is different from research that instead of testing for all

2
00:00:20.180 --> 00:00:21.390
parameter combinations,

3
00:00:22.170 --> 00:00:27.030
we specify distributions in our parameter grid and randomized search

4
00:00:27.050 --> 00:00:31.820
and randomly samples values from these distributions. Uh, you'll know,

5
00:00:32.000 --> 00:00:36.500
uh, what I mean on this example. Uh, here I define, uh,

6
00:00:36.500 --> 00:00:40.950
three distributions. First is discrete uniform distribution,

7
00:00:41.080 --> 00:00:42.750
generating random integers.

8
00:00:43.850 --> 00:00:46.670
Second is uniform continuous distribution,

9
00:00:47.370 --> 00:00:50.190
and the third one is exponential distribution.

10
00:00:52.390 --> 00:00:57.220
Their parameters in the brackets are minimum and maximum value. Uh, we can, uh,

11
00:00:57.220 --> 00:00:59.660
specify of course any other type of distribution.

12
00:01:02.080 --> 00:01:06.630
Using the r v s method. We randomly sample a value from this distribution.

13
00:01:07.170 --> 00:01:11.310
And when we repeat that, for example, 1000 times, like in this loop,

14
00:01:12.750 --> 00:01:17.410
we can plot then the histogram of this distribution and its shape is now

15
00:01:17.410 --> 00:01:18.410
clearly to see

16
00:01:26.010 --> 00:01:30.780
this way we define our parameter grid, and in the randomized search,

17
00:01:30.860 --> 00:01:35.180
cross validation, uh, we have to specify also, uh,

18
00:01:35.180 --> 00:01:37.540
number of iterations, uh, parameter,

19
00:01:37.870 --> 00:01:42.290
which limits the number of parameter combinations to test other

20
00:01:42.300 --> 00:01:45.890
parameters are same as in grid search. In

21
00:01:50.280 --> 00:01:54.800
our next video, we will learn more about tune search, cross validation.
