5 minutes with you to create models and discriminant model
What is the discriminant model / Discriminative Models
We should first do not fix those arcane mathematical formulas, to cite two examples
An example of a
Suppose a phone has three characteristics, our task is to predict a consumer will buy or not buy the phone.
So discriminant model is based on some prior knowledge already know, from buying and not enough to buy a selection of these two possibilities.
| feature | Influence on decision-making |
|---|---|
| Running smoothly | Purchase possibilities +0.1 |
| Charging slowly | Do not buy the possibility of +0.4 |
| 5.5-inch screen | Unchanged (this can be considered consumers do not care about the size of the phone) |
Examples of the dicarboxylic
Digital Recognition: given a map, there are a handwritten digits, requires several machine identification number is, 0 to 9, then the decision of 10 possible. In recognition of the picture when the computer is also little pieces do convolution extract features, then for the convenience of explanation, here I have manually extracted three adults can read and restore features of the way, as follows.
| feature | Influence on decision-making |
|---|---|
Features a
|
It is to increase the probability of numbers 1,4 |
Features two
|
Increasing numbers of probability 2,4,3,6 |
Three characteristics
|
It is to increase the probability of numbers 2,0,8,3,9 |
Whatever the example, it is some of the information has been obtained by the auxiliary judge, given a final decision. That's why those articles are filled with formulas called conditional probabilities.
In summary, discriminant model is used to predict the probability of an event has occurred under the conditions of y, x events.
Specific examples above, under conditions wherein a is occurring, the probability is the number 4. Or already know when the phone is charging slowly, the probability of buying mobile phones.
What is the generative model Generative Models
| Conditional Probability | Joint probability |
|---|---|
| Discriminant model | Generation model |
Having a conditional probability, the next step is the joint probability. Discriminant model learning of conditional probability, generated model of learning is the joint probability.
In order to better compare the difference, we use the example above:
An example of a
Here we want to change a task, in order to generate a model to understand. Our task is to do market research, and complete the following form.
| Number of incidents | buy | No purchase |
|---|---|---|
| Running smoothly | 220 | 30 |
| Charging slowly | 19 | 100 |
| 5.5-inch screen | 45 | 67 |
So this time we need to look to enlarge the market. We need to get the phone running smoothly and was purchased by simultaneous joint probability, of course, there are many such combinations, such as mobile phones 5.5 inch not purchased, charging the phone is slow to buy more.
Examples of the dicarboxylic
Similarly, where we need to give figures were present and wherein a probability of 0 to 9, and features of the two numbers are present and the probability of 0 to 9.
If all of these handwritten numbers from one hand, we have figured out the features, and even be used to generate this person's handwriting, because this man to write out what features which words, what is the probability that we all know.
Sentence summary
Discriminant model in your father at the height of more than 180 known conditions, forecast your height will not exceed 180. If your father is higher than 180, then you will increase the probability of higher than 180. But the height of the probability distribution around the world who temporarily has not changed.
(For scientific rigor, where the father has been changed to a father and not his father)
Generation model is random after you gave the highest height of an adult. The probability is much taller than 180, this probability can only be decided in accordance with the frequency distribution of the height of all adults worldwide.