This wordpress site allows highlighting code and rendering equations written in LaTeX.
Example of syntax highlighting in R:
<pre lang=”rsplus”>
a=4 # assigning value of 4 to a
a # print value of a
</pre>
Example of equation using LaTeX code:
Example of syntax highlighting in R:
a=4 # assigning value of 4 to a
a # print value of a
Professor and/or TA:
Just for clarification, am I correct in saying that we may submit the service requirement through this Friday (12/09) ? The assignment page says the Fri 12/07. But the 7th is Wednesday.
@cfu3: It should be Friday Dec. 09 11:55pm.
@all: HW4 grades are posted.
Professor, TAs:
Have you decided yet what the cut-offs will be for A, B, C, etc. will be?
Thanks
I have got it wrong for the relationship of Fisher information. Fisher information is equal to the variance matrix of the score vectors thus always positive semi definite. However, the Fisher information equals to the EXPECTATION of the negative Hessian, not the Hessian itself. therefore, we cannot claim anything for the Hessian.
Also, will the final exam be like the midterm, in that we are given a list of questions and told to pick a certain number of them to answer?
@zixu1986: yes you are right. The Fisher information is the expectation of the Hessian of the log-likelihood.
@dlong: I don’t know yet how many questions there will be. The cut off points will be determined after seeing all the grades. I will try to be generous.
@all: just posted link to slides and the book I mentioned that has the trees content.
Dear Professor,
Is the final exam open book ? ( sorry if the question has been already answered before)
professor or TA, I can see the scores of hw4 and hw5 under ‘assignment’ on T-square, but I cannot see them in ‘gradebook’. Did TA release the scores to us?
Prof. Lebanon: What is the range of exam 2? Does it cover the contents we learned before exam 1?
professor or TAs, will you post the notes of the second and third extra classes? I couldn’t be there because of time conflict.
The videos of all 3 extra classes are here: http://smlv.cc.gatech.edu/cse6740-lecture-videos/
Is stagewise feature selection with replacement equivalent to stepwise feature selection with replacement?
Does anyone know where the exam will be held? Is it the same classroom?
well, I have emailed the professor and he has replied that it is open book and covers everything since the midterm.
@all: the exam is open books, open notes. You may bring everything you want except for computers, cell-phones, calculators, internet devices, etc. It will cover material that was not covered in the midterm. For time information see http://www.registrar.gatech.edu/students/calendar.php.
@rohitbanga: no.
@all: please make sure your grades for all the assignments that you completed including the service requirements are available on t-square (give the TA another couple of days to grade the service requirement). If that is not the case, email the TA.
Dear Prof Lebanon,
I am repeating zixu1986′s question as it has not been answered yet.
What is the venue for the final exam? Is it same as the class?
Thanks!
In the case of non-linear SVM and dual problem of linear SVM, the classification equations needs training data points as well. Why do we need training data even though SVM is parametric?
@Shruti: it seems so. However, deciding on the time and place is done by GT and not by me.
@ramakicha: I don’t understand the question. Both parametric and non-parametric methods need training data. And of course SVM does also (both dual and primal optimization problems use a training data).
Will we be given choices in the exam like the midterm test ?
In the case of non-linear SVM and dual problem of linear SVM, the classification equations needs training data points as well. Do we need training data for “classifying the test data point”?
@Gunboy mine is out on “assignments” but not under “gradebook”.
@ramakicha: If I understand you correctly, you are wondering about classifying future points once an SVM has been trained and whether the training data is needed for this step.
For a linear SVM, the weight vector w corresponding to the separating hyperplane is computed from the training data, but after that the training data is no longer needed. Classifying a point x can be done using the inner product of w and x.
For non-linear SVM, you are correct that the training data is needed in order to classify a point, but only the support vectors are needed. One hopes that the number of support vectors is small compared to the size of the entire training set. The reason the support vectors are needed is because the classification is being done with respect to a hyperplane in a higher dimensional space, and the weight vector for that hyperplane does not in general correspond to a single vector in the original space.
@ramakicha: this has to do with whether you solve the primal or the dual problem. If you use the primal SVM formulation, then you get the w vector and then you can discard the training data. If you use the dual SVM formulation then the training phase concludes with the Lagrange multipliers, but in order to get a classification rule you still need the training data (at least the support vectors).
This is the same for either linear or non-linear SVM, only that for non-linear SVM the primal formulation is very rarely used.
Can somebody help me out with the answer for following question..
Show a training set for Logistic regression where
a. convergence to global maximum occurs …
b. no convergence occurs…
c. weights tends to inf
d. weights tends to – inf ….
@Gunboy I found an article that might explain a and b. Particularly, scroll to “what can go wrong” for the answer to b.
i stupidly forgot to paste the link to the article. Sorry about that.
http://www2.sas.com/proceedings/forum2008/360-2008.pdf
You must be logged in to post a comment.
