WEBVTT - This file was automatically generated by VIMEO

0
00:00:07.830 --> 00:00:12.330
Hi. In this video we will see a few other interesting prototypes,

1
00:00:12.620 --> 00:00:17.130
which can give us more information about underlying patterns and visualize

2
00:00:17.280 --> 00:00:18.930
information more effectively.

3
00:00:21.220 --> 00:00:24.760
We have already tried before the one dimensional density plot,

4
00:00:24.940 --> 00:00:28.240
but this chart can be done also using two variables.

5
00:00:29.160 --> 00:00:33.940
We see it here in this first example where now we see clearly

6
00:00:34.290 --> 00:00:38.620
what are the most dense areas when analyzing the relationship between count and

7
00:00:38.620 --> 00:00:39.740
service count variables.

8
00:00:42.940 --> 00:00:47.350
Here we can see the density when filtering only normal connections with values

9
00:00:47.440 --> 00:00:51.350
under 50. We sample the data here, uh,

10
00:00:51.490 --> 00:00:56.390
and we did it also in the previous example because the density for

11
00:00:56.390 --> 00:00:59.430
whole data set takes, uh, while to calculate

12
00:01:02.080 --> 00:01:06.970
similar information we can get with the scatter plot histogram using

13
00:01:07.230 --> 00:01:08.810
sea born's, uh, joint plot.

14
00:01:09.820 --> 00:01:13.320
But we can see the distribution of both variables here

15
00:01:17.590 --> 00:01:22.490
in, uh, one of our previous modules we used already the scatter plot from

16
00:01:22.530 --> 00:01:26.550
seaborne with added regression line, just as a reminder.

17
00:01:30.210 --> 00:01:34.910
But the body is also more interesting and it helps also a lot

18
00:01:35.730 --> 00:01:38.350
is the lowest curve. Uh, that is, uh,

19
00:01:38.350 --> 00:01:42.660
basically a locally elevated linear regression. Here,

20
00:01:43.000 --> 00:01:47.700
it seems like a straight line, uh, but can filter some data out.

21
00:01:47.880 --> 00:01:51.660
And uh, we see that it's basically a curve.

22
00:01:52.650 --> 00:01:57.550
We can see basically much more clearly the extent of the linear relationship

23
00:01:57.550 --> 00:01:58.390
of the variables.

24
00:02:01.720 --> 00:02:05.050
Another useful chart type is the tree map,

25
00:02:05.180 --> 00:02:09.410
which can be used to better see the counts of multiple categorical variables.

26
00:02:10.170 --> 00:02:10.530
Uh,

27
00:02:10.530 --> 00:02:15.350
but we can also map numeric variables to bin size or color as well.

28
00:02:18.660 --> 00:02:23.240
The violin plots are combination of, uh, density plot and a box plot.

29
00:02:26.850 --> 00:02:30.350
And the last chart type, which are use very frequently,

30
00:02:31.210 --> 00:02:36.020
is the other plot. They are very useful, uh, when making, uh,

31
00:02:36.090 --> 00:02:40.070
profiling of some groups or clusters, for example. Uh,

32
00:02:40.070 --> 00:02:44.830
basically the numeric variables are plotted in a circle where each line is

33
00:02:44.990 --> 00:02:45.870
a specific group.

34
00:02:51.460 --> 00:02:55.620
I usually also pre-process the data with standard scaler,

35
00:02:55.790 --> 00:03:00.160
where zero is the average and one is the standard deviation. Uh,

36
00:03:00.260 --> 00:03:04.360
but you can use also other scaling methods like, uh,

37
00:03:04.380 --> 00:03:09.000
robust scaler, for example, where you do not use average and standard deviation,

38
00:03:09.220 --> 00:03:10.120
but quantiles,

39
00:03:14.570 --> 00:03:18.870
we have to also reshape the data from wide to long format, uh,

40
00:03:18.870 --> 00:03:21.190
using melt method from pandas.

41
00:03:24.470 --> 00:03:27.410
And then we use the line polar method from Plotly.

42
00:03:31.720 --> 00:03:35.220
In our next video, we will check two Jupiter lab extensions,

43
00:03:35.230 --> 00:03:39.260
which help us a little with the plotting and exploratory analysis.
