WEBVTT 00:01.200 --> 00:08.310 Now that we have the both unsupervised learning, let us have a discussion on different types of evaluation 00:08.310 --> 00:13.230 parameters, which we have so clustering is used for. 00:14.790 --> 00:17.820 Grouping different types of data, datapoint. 00:18.210 --> 00:24.000 Now here you can see we have different types of data points present and these have been clustered by 00:24.000 --> 00:25.690 different methods which represent. 00:27.120 --> 00:32.430 These different methods allow to create clusters with different properties. 00:33.450 --> 00:41.700 Now, this particular documentation, which is present inside Gitlow, provide a complete overview of 00:41.700 --> 00:44.410 how each algorithm can be implemented. 00:45.030 --> 00:53.730 So when we go below, we can directly see the implementations like DV scan, hierarchical clustering. 00:56.820 --> 00:57.780 Gaming's. 00:59.480 --> 01:00.470 And rescan. 01:01.630 --> 01:09.130 All of these have been implemented here, and you can see the implementation and details about the same 01:09.130 --> 01:11.830 by going inside the demo for the same. 01:19.490 --> 01:22.190 Here is the entire implementation of this. 01:26.310 --> 01:33.060 This is another documentation for different types of examples, which we have, so here you can actually 01:33.060 --> 01:40.620 find out different examples for classification problems, flustering problems for convenience, estimation. 01:41.630 --> 01:48.410 Different data set examples, decision trees, ensemble methods and different methods which are available, 01:48.710 --> 01:51.620 and there are different exercises also which are available. 01:51.630 --> 01:56.690 So you can try these out and see how good you can work on these. 01:57.220 --> 02:01.890 Next, we will try to have a look at the demonstrations. 02:01.930 --> 02:09.200 So here we have different demonstrations, all of gaming's and different type of algorithms which are 02:09.200 --> 02:09.660 present. 02:10.460 --> 02:16.040 So here we can see the hudud, the clustering, the underground, which has been created here. 02:17.110 --> 02:19.120 And how we can plot this. 02:23.890 --> 02:25.800 Again, you can go to plastering. 02:26.960 --> 02:28.520 And see the. 02:29.940 --> 02:32.310 Implementation of gaming's here. 02:33.500 --> 02:38.930 This has the complete implementation of gaming's with the plot. 02:44.180 --> 02:46.340 For those who can again go up. 02:48.990 --> 02:55.170 And again, have a look at a different clustering algorithms, the demos, the examples, so it has 02:55.170 --> 02:57.680 a lot of information which you can use. 02:57.690 --> 03:01.310 And here is a comparing different clusters and algorithms. 03:01.590 --> 03:05.790 So a lot of examples are present here for your exploration. 03:06.480 --> 03:11.930 This is the website for Psychic Land, which has all these information available. 03:13.430 --> 03:20.210 Then you can go to the user guide, which is present above this, you visit guide contains details of 03:20.210 --> 03:28.720 all the supervised learning algorithms, unsupervised learning algorithms, how we can do modern selection, 03:28.730 --> 03:30.940 how we can evaluate the models. 03:31.980 --> 03:39.240 Lake cross-validation during the hypovolemic, those metrics and scoring model, persistance, validation 03:39.370 --> 03:39.720 of. 03:41.400 --> 03:50.100 How we can do that transformations how we can load different detours so all of these things are given 03:50.100 --> 03:57.100 in this particular Whipsnade, which is very nice and very nicely explained, everything is provided. 03:58.020 --> 04:00.580 So we'll go to clustering here and clustering. 04:00.720 --> 04:05.640 We have this last option, which is clustering performance evaluations, which we go through this. 04:06.300 --> 04:13.530 So we have several metrics which are present for clustering performance evaluation, first one being 04:13.530 --> 04:15.960 the adjusted right index. 04:17.600 --> 04:26.480 It is the for the ground truth, for the class assignment, so are clustering algorithm of language 04:26.480 --> 04:33.590 of the same level and just an index is the function that measures the similarity of the assignments, 04:33.590 --> 04:37.170 ignoring full mutation and with chance normalization. 04:37.640 --> 04:45.080 So it will basically compare the actually deals with the labels, which we have provided and tell us 04:45.260 --> 04:47.090 how good our clustering is. 04:47.690 --> 04:51.560 But this is used when we already have the labels of. 04:53.010 --> 05:00.870 So the advantages are that this particular labelled assignment having at a school close to zero point 05:00.900 --> 05:10.550 zero, so for any value of interested in sample, which is not the case for index and the measure, 05:10.860 --> 05:12.390 so these can be used. 05:13.630 --> 05:23.620 Then there is another index, which is mutual mutual information based school, which can again be found 05:24.070 --> 05:28.980 using Escalon metric DOT adjusted mutual information. 05:29.290 --> 05:31.990 So this is the school which you can actually use. 05:35.350 --> 05:38.200 The perfect labeling school would be one. 05:40.760 --> 05:49.790 Next, which we have, is the homogeneity or completeness and we measure these are different measures 05:50.150 --> 05:50.910 which we have. 05:51.310 --> 05:58.850 So this has the concept of homogeneity, score and completeness for so homogeneity is what gives how 05:58.850 --> 06:01.820 homogeneous the data is, the clusters. 06:05.160 --> 06:17.340 So we can find this out, and the one which is most important and mostly used is the psyllid coefficient 06:17.940 --> 06:24.990 psyllid coefficient is of comparing the mean distance between a sample and all other points in the same 06:24.990 --> 06:33.480 cluster that the mean distance between the sample and all other points belonging to the next nearest 06:33.480 --> 06:34.050 clusters. 06:34.440 --> 06:41.520 So based on these distances, that is the inverted cluster distance and the cluster distance. 06:41.790 --> 06:48.540 It finds out this and it's called the Senate score ranges from minus one to one there. 06:48.810 --> 06:52.440 One is the perfect cluster formation. 06:52.770 --> 07:00.060 Negative one means that the cluster has been misclassified and the zettl means that the clusters are 07:00.060 --> 07:01.420 overlapping in nature. 07:01.690 --> 07:05.460 And here you can again find the implementation of the same. 07:05.880 --> 07:12.480 We will be using this implementation while we will be implementing different models in our lecture's. 07:13.660 --> 07:22.270 So the simple implementation is we will involved the skill on import metric and from the metric, we 07:22.270 --> 07:25.500 will simply use metric dorsolateral. 07:25.930 --> 07:33.340 We will give the values, the labels and the metric that you want to use for the distance calculation, 07:33.700 --> 07:37.130 and it will give the values which should be near two one. 07:37.840 --> 07:45.810 So this is one very important index and most importantly used index, which is political efficient. 07:46.060 --> 07:52.120 So this is what we will be using for this particular obsession and for all the. 07:53.560 --> 07:56.950 Of clustering algorithms, which we will be implementing. 07:57.490 --> 08:07.210 So this is about the Escalon Library and how you can go through this library and take different algorithms 08:07.210 --> 08:10.540 and the examples which are available under this. 08:11.080 --> 08:18.340 So you can go to the examples and see different examples available for all types of problems which are 08:18.340 --> 08:18.850 present. 08:20.130 --> 08:27.510 Similarly, you can go to the user guide and under User Guide, you can look up any algorithm which 08:27.510 --> 08:35.830 you want to find about and learn about it next in case you want to learn about the EPA itself. 08:35.850 --> 08:44.820 You can simply go to the EPA and it will start showing the the methods and different usually used libraries 08:44.970 --> 08:46.590 from this particular EPA. 08:48.390 --> 08:57.020 So this is about Escalon and how we can actually use different coefficients and most importantly, Sillett 08:57.060 --> 09:00.810 index for finding out the goodness of a cluster. 09:01.350 --> 09:02.710 So thank you. 09:02.880 --> 09:10.380 In the next session, we will be learning about Haralson clustering and then we will talk about different 09:10.380 --> 09:12.360 clustering methods which are available.