Hello,
I have done several experiments. For each one, I have a couple (xi, yi) where xi is the label and yi is a measurement. However for the same input value (xi) I can get different values for the output (yi) provided by different tests. I need to use a machine learning method to model the problem and then predict the measurement yj for a given xj.
I have multiple questions:
- Is the problem well defined?
- What is the method that can I use?
Thank you,
Hello,if i understood your question well,it seems for as a classification problem .But i am not sure since you said that Xi are the label since in most of the case Yi (output) are the more of labels or classes.. If that is the case, regression model can be used.. if you have so many features(Xi=ax1+bx2+cx3+dx4+.....) where x1 ,x3,x4 ,...xi have big dimensionality( can have lots of categorical or continous values ) regression model doesnt work well....I would recommend using Gradiant boosting based on Oblivious trees
http://wan.poly.edu/KDD2012/forms/workshop/ADKDD12/doc/a3.pdf
Hi Mariem,
What values can Yi take? is it discrete or not?
Do you wish to obtain a prediction of Yj for each of the tests?
Hello,
Thank you for your response.
I have the following couple:(state, reliability). where reliability is a number between 0 and 1.
I have done several experiments in each one I compute a couple of values (state, reliability). But, for the same state (in different tests) I have different values for reliability.
For example:
(state = 5, reliability) :{ (5, 0.1), (5, 0.2), (5, 0.2), (5, 0.6)}
(state = 10, reliability) : {(10, 0.3, 0.4, 0.6})
.....
So, I don't think that regression can work. I was wondering which is the best machine learning to use in this case. I think that bayesian approach can be used. What do you think?
@ Elie Magambo Gatete, I download the file and I will examine it.
@ Shahar Bar, Yi is not discrete but is in [0,1]. Yes, I want a prediction of Yi for a given state.
Thank you
The fact you are getting different Y values for the same values of X is an indication for a measurement noise I guess. I think you can use regression, by averaging over the various Y values coming from the same X value to get a noise reduced single Y value for each X value.
Maybe you can show us a plot (X,Y) so that we can see the whether it is a regression or classification problem? Thanks.
Thank you Linh,
Here a sheet that contains results of experiments. For each value of X, I have done 100 experiments so I got values of Y.
The goal is: for a given value of X (even unobserved value) I can estimate the value of Y.
Thank you very much,
I guess you want to estimate y_i=f(x_i), but since you have only one training x with multiple choices of y, it's weird though. The best I can guest (without prior knowledge about the underlined experiment) is that, given x, I can give you the average of y. Machine Learning probably can't help...
Hello Linh,
I did not understand, you say that I have only one training x. What does it mean?
I have done several experiments, in each one x have a value.
Thanks
Could you try to check the correlation between x and y? I have check, and seem like it is very low (close to zero), so I would think there is no hope to find function f such that y=f(x).
How do you obtain the value of X? Is it just an arbitrary identifier/tag, or does it have some sort of (perhaps physical) meaning?
Also, do you have more information about each state X? If it has several different features (that is, if X is a vector with multiple components, or dimension), the problem would be closer to a traditional machine learning problem.
Thank you Rodrigo de Salvo Braz,
Values of X given in the file are some samples. They indicate a geometric transformation between two images. So it is a unique information and I don't think it has features. What do you think?
Still not clear to me what you mean. How exactly do you obtain the value of X? Where do these numbers come from?
Rodrigo,
X is a known parameter of a camera, it can have values between 0 and 355.
You can see that X and Y are uncorrelated, so it's not possible to predict I think.
Yes, given that X is a camera's parameter and there is no correlation, it would seem so.
I looked at the spreadsheet of data you provided.
While I don't think this is what you want to hear, the simple idea of plotting all the Y series for the X values shows that many of these Y series have identifiable trends / patterns, but they seem to be all quite different. Is there any reason to expect that the Y values would vary in a systematic way. How about taking "differences" --- so plot X vs (Y1 - Yn). This seems to be a problem where basic data exploration as opposed to machine learning is going to be useful to you.
- Badges
- Science topic