I do not know how to do in the below case in Naive Bayes.
I want to predict probability of each class by using continuous variables.
In my case, class's name are A,B,C,D and prior probability is same(0.25).
And Data which I have is as below.
V1 V2 V3 V4 Class
0.25 0.12 0.30 0.21 A
0.42 0.53 0.28 0.19 B
0.21 0.34 0.33 0.44 C
0.13 0.23 0.34 0.55 A
・
・
・
In my case, class is four and is hypothesized to have same prior probability, but Data does not include one class (D).
When one data(0.21,0.32,0.44,0.11) is given, can I predict each class's probability, especially D?
I am beginner in bayes theory, but I believe that I can get that value after Bayesian update.
If possible, please tell me the package or library (R or python is better for me and other language is OK).
I don't believe you can do this: predict the probability of one inexisting class. Is, more or less, classify a human as martian whitout to know how a martian is.
Hi Ryosaku,
In the bellow reference you can find a beautiful demonstration how you can go from a real to a probabilistic unit space mapping using as reference the softmax transformation. In this section Ian Goodfellow and collaborators proof how you can build a probability distribution from a softmax transformation on continuous variables under quite acceptable assumptions. The softmax function can definitely help you in this problem.
If the reply was useful, please, recommend for others,
Reference (page 178, section 6.2.2.3 - Softmax units for multinomial output distribution):
https://www.amazon.com/Deep-Learning-Adaptive-Computation-Machine/dp/0262035618/ref=asc_df_0262035618/?tag=hyprod-20&linkCode=df0&hvadid=312128454859&hvpos=1o1&hvnetw=g&hvrand=9218063798699755148&hvpone=&hvptwo=&hvqmt=&hvdev=c&hvdvcmdl=&hvlocint=&hvlocphy=9005781&hvtargid=pla-416263148149&psc=1
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Programming Languages
- Python