Künstliche Intelligenz und Maschinelles Lernen in der Praxis

1. 00:00Now that we have our data in our program, we can now
2. 00:04Simply follow this data analyze and visualize.
3. 00:08That just helps us understand what we're doing. and we're going to use it for data
4. 00:12process. Even though we are at the end of Apply a machine learning model daily,
5. 00:16is nevertheless important, that we understand the data.
6. 00:21And we can now do that with our data. just start because we have to measure the longitude and the
7. 00:26Latitude, which means that it's ultimately Geodata, to visualize this very nicely.
8. 00:33We can make this latitude and longitude simple visualize in a so-called scatter plot.
9. 00:39That is, the data is represented like this. on the x-axis Longitude and
10. 00:46on Y-axis Latitude.
11. 00:48And we can already see. approximately how the residential areas are distributed.
12. 00:54But that doesn't really help us much.
13. 00:56That is why we are enriching them Visualize step by step.
14. 01:01And the next step that we do well is simply to these individual points
15. 01:05information, for example on the Population, i.e. population density
16. 01:10or the house value.
17. 01:12This is also relatively easy.
18. 01:14We can visualize the data and see it. now in our visualization not only where this
19. 01:20residential areas, but also, how much they cost and how many
20. 01:25People live in this neighborhood.
21. 01:27And here you can see quite well that here at the course of this edge apparently more expensive
22. 01:34Apartments are more like up here.
23. 01:37That might help us now very good.
24. 01:40However, we can take another step and then visualize the data even better.
25. 01:44Now, if we know that's geospatial data from California, we can now
26. 01:49see that we have a picture of Find California in those places
27. 01:54and finally deposit it.
28. 01:56And if you do that, you know, this picture. , we can visualize that very easily.
29. 02:04And I have a little script here before which we import again here.
30. 02:09And if we do it now visualize, we see now where on the
31. 02:15Map these residential areas are accurate.
32. 02:18Yeah, well, now we can really gradually to better understand why they
33. 02:23Neighborhoods down here on the Edge are more expensive than the top.
34. 02:26It is probably simply because they are closer by the sea and by the sea the apartments are more expensive.
35. 02:33This could be the first be good or a first good element,
36. 02:38what we understand about our data: The closer the residential areas by the sea are the more expensive.
37. 02:45So, for example, what we've just seen is this. I'm going to take another step back.
38. 02:50that expensive neighborhoods seem to even are more often grouped, i.e. are close to each other.
39. 02:56Whatever normal, a information.
40. 03:00So now we have two pieces of information. that we were able to derive from our visualization.
41. 03:05Now we know that and we can use it go even deeper into our analysis and
42. 03:11better understand our data. What we can just do it now.
43. 03:14when we look at our data and we know that which attributes we have, then we can
44. 03:18Maybe come to the point that ocean proximity, which translates as proximity to the sea
45. 03:25And I want you to look at these values and see now. that in our data frame the categorical values
46. 03:31, such as less than an hour to the sea or a domestic.
47. 03:39Yeah, so how this houses categorized are able to understand that now
48. 03:44And maybe work pretty well on that.
49. 03:46But we're also looking directly at this. that the expression Iceland, i.e. island,
50. 03:53five times and we have to I'd like to start thinking about this at once.
51. 03:59What we can do well now is simple to group these values
52. 04:05and the median value.
53. 04:07And that's where we can see something straight out there. very exciting what we also just
54. 04:10in our data.
55. 04:12And that is that the media price on this grouped data, if we just
56. 04:17on our data, see that we have significantly lower prices in Germany
57. 04:24as near to sea, i.e. less than an hour,
58. 04:27or close to a bay or with of the distinctions right next to the sea.
59. 04:31So, you have to look again in the data, as now close to the sea
60. 04:34and under an hour.
61. 04:36Maybe near the sea something like this like five or ten minutes to the sea.
62. 04:40But we're already seeing relatively Well, that's kind of different.
63. 04:44The value Iceland has the greatest value, but again
64. 04:50Caution, there are five points on that we measured.
65. 04:53Whether this is particularly meaningful now , we have to ask questions.
66. 04:58And what's going to do very nicely? I'd like to take a moment to review that.
67. 05:03is this visualization in stages To look at us, that is to say
68. 05:08on a filtered data set according to the Ocean Proximity.
69. 05:12And for that, just write us a little Help function that matches our data frame exactly
70. 05:17filters and can now be incremented rebuild our visualization.
71. 05:23We're going to start with just making the Ads that have as category Domestic.
72. 05:28When we do that, we see it. are actually almost just blue dots.
73. 05:35And blue on our color scale means that it really simple cheap values, cheap houses.
74. 05:43We can do the same for under for one hour to the ocean.
75. 05:46See, here are some more blue, but also more in the A red colors
76. 05:54, which again means more expensive prices.
77. 05:57And the same for near Ocean.
78. 06:00Wherever we see, it will It seems to be getting more and more red.
79. 06:05And so our prices are always more expensive, just for understanding.
80. 06:11We can use this visualization to get more and more our data and also sometimes
81. 06:17To play with it, to look gradually, how you can build visualizations like that.
82. 06:23And they have another The idea that the closer a house is to the sea,
83. 06:29The more expensive it becomes again somewhere.
84. 06:32Let's go back to the two that we've created. still outstanding. This is near bay once.
85. 06:38And here we see, apparently, that there is a specific bay where all the houses are
86. 06:43also collect. That is, this seems to be the bay to be mentioned in our data.
87. 06:49And as just announced we still have Iceland as an outlier.
88. 06:55And when we plot that data, We need to look very carefully.
89. 07:01where the points are now and see That seems to be where our data is.
90. 07:08Now we really have to think, is enough for us to make any statement about a house,
91. 07:12What's on an Iceland?
92. 07:14Because apparently the prices so different,
93. 07:17That means that we may already have a first And the evidence is that we may be looking at this data.
94. 07:21Do not want to use training because they Maybe we'll distort our training.
95. 07:26perhaps our solution to hybrid solution.
96. 07:28So for neighborhoods where we have enough data. quietly use our model,n but for data that
97. 07:36So houses that are in Iceland, maybe but would rather ask the person again.
98. 07:42So we can start right here, and start thinking how we want to deal with it.
99. 07:47And for now we just want to Let's make the decision that we're going to take Iceland.
100. 07:54So islands, where we really only have five values. that we just remove them from our data.
101. 07:59We don't have enough of what we know about it. and don't want to use that for our training.
102. 08:04That is now a possible decision.
103. 08:10What we are now living from our experience is that one is theoretical
104. 08:15already somewhere a forecast which could be based purely on the
105. 08:19Proximity to the sea is based, that is has nothing to do with machine learning.
106. 08:23So we're not at our model yet, but we're which you could use anyway.
107. 08:28And we just want to try that. by taking the constant value per group.
108. 08:33For example, as soon as is a house in the interior,
109. 08:35we just get the median price for domestic houses.
110. 08:38We can just apply that and on our data.
111. 08:43And if you scroll a little bit to the right, We can counter this and see that we can
112. 08:48perhaps still strongly do, like For example, a domestic house.
113. 08:52Now we have one here, which seems to be is significantly more expensive than our median.
114. 08:56That is, we would have a They're going to make a very bad forecast.
115. 08:59But nonetheless, it's just about to play around with the data, to hypothesize
116. 09:05and maybe just a baseline, So you have to build a foundation against which you can then
117. 09:09and can always be better.
118. 09:14Exactly, here again held, so you can create a kind of a bassline like that.
119. 09:19can't really hurt.
120. 09:20Because then you already It's something to build on.
121. 09:26What we want to look at now? before we go to the AI training, the
122. 09:30Correlation analysis, that is, we have now our categorical data, which is proximity to the sea.
123. 09:35information. And now we want to look at what it looks like on numeric data.
124. 09:39You can see what's called the Pearson. Calculate correlation coefficients.
125. 09:44This is one of several possible, and And that's what you're going to try to understand
126. 09:50if we have a target variable, a target variable Y, for example,
127. 09:55and we have other variables X if X increases or decreases by a certain value,
128. 10:02how does this affect our target variable Y?
129. 10:05And the higher these correlation coefficients, So now in amount, the bigger the effect.
130. 10:11What you have to watch out for , this is only for a linear context.
131. 10:16We are looking at it now anyway but we For example, you can see very quickly.
132. 10:22yes also somewhere sense, the more people in earn the living area, the more expensive
133. 10:27So on average, these are the neighborhoods.
134. 10:29Because then you just might be in the in a more expensive residential area.
135. 10:35We can also do that and visualize.
136. 10:39And you can see there, so it's on the x-axis. the median income in thousands
137. 10:45and Y-axis our target variable.
138. 10:47And again, you can see It's certainly a trend.
139. 10:49But you have to recognize it. that there's just one big scatter.
140. 10:54that means we don't have a perfect explanation from our Data on income.
141. 10:59What we see again here, the we now find again and again in our data
142. 11:04if we plot our target variable, the up here at 500,000, you get a lot of values.
143. 11:09This comes via this cut, the we have already established before.
144. 11:12So we kind of have to think about this. really want to put this project into practice
145. 11:18how to deal with such values.
146. 11:22Exactly, even this again held here.
147. 11:25So the more people earn, the higher the price can be.
148. 11:29But above all, that at 500,000 euros appears to have been set an upper limit.
149. 11:35And we now choose this In case we keep the data, just
150. 11:40take as they are, but might have to take a practical project consider whether this is different
151. 11:44should be addressed.
152. 11:46Now we can still consider whether we can the attributes that we have that are somehow
153. 11:52may explain a correlation. to build additional attributes.
154. 11:55So for example, we could totally make bedrooms. sharing through total rooms
155. 12:00the share of bedrooms.
156. 12:02Maybe it is with houses that have more bedrooms than in proportion,
157. 12:07these may be more expensive apartments.
158. 12:09Or people per household.
159. 12:13So we can still have operations on our data. to develop additional attributes and
160. 12:17test and watch this time, whether you can work with it.
161. 12:20And we can do that in a very simple way. calculate our data and then again
162. 12:27display the correlation vector to us.
163. 12:31And recognize here, for example, that seemingly ratio Bedrooms a relatively good
164. 12:38has a correlation value that we can at least You can use it and try it.
165. 12:42Maybe our AI model works well.
166. 12:45That means here too, the data is simply try a little, play around and watch
167. 12:50what could it do for even more good Provide information for our model?
168. 12:55And that's exactly what we have our data with relatively nicely analyzed, have a quite good
169. 13:00image of our data and can now be with relatively good confidence
170. 13:05our first AI model in the next video.

To enable the transcript, please select a language in the video player settings menu.