Druss2k

Germany

Jul
15
comment How are piecewise cubic spline bases constructed?
If I look at $N_{i,0} = \begin{cases} 1, t\in [t_i,t_{i+1})\\ 0, \text{else}\end{cases}$ (which is a first order B-Spline) and integrate three times, I'll get $N_{i,0} =\frac{1}{6}t^3+\frac{1}{2}p_1+p_2t+p_3$ where $p_i$ are constants of integration from each step. I do not quite understand how you'll get $\frac{1}{3!}(x-\xi)^3_+$. Additionally to get a smooth, twice differentiable cublic spline function I would need to impose conditions on the first and second derivative. Looking at statistical text-books such as Wood - GAM in R, I dont quite get how a cubic spline basis actually looks like.
Jul
11
asked How does the cubic spline basis exactly look like
Jul
11
comment Parallel Design-Matrix, Parameter-Vector multiplication in ML-Estimation
This is what I did but unfortunately things did not look very promising. Well it was worth a try :-)
Jul
11
comment Parallel Design-Matrix, Parameter-Vector multiplication in ML-Estimation
To comment on your assumptions and the real problem. The evaluation of the likelihood/gradient is basically nothing else then matrix multiplication done multiple times. I will need to evaluate those functions ~200-300times. Each time, the unchanged design-matrix gets postmultiplied by a changing vector. Hence your suggestion does fit the problem. The reason why I thought I could improve my performace is that the matrix has about 3mio rows. So splitting the job up into 3 multiplications of 1mio rows and then just returning the result should usually be faster?
Jul
11
comment Parallel Design-Matrix, Parameter-Vector multiplication in ML-Estimation
I edited my previous post to inlude your function in my comparison. Please have a look down from the line /edit
Jul
11
revised Parallel Design-Matrix, Parameter-Vector multiplication in ML-Estimation
added 1144 characters in body
Jul
11
accepted Parallel Design-Matrix, Parameter-Vector multiplication in ML-Estimation
Jul
11
awarded Curious
Jul
10
asked Parallel Design-Matrix, Parameter-Vector multiplication in ML-Estimation
Jul
7
comment What is the simplest way to classify airplan manuvers?
Ive never worked with DTW but my suggestions is based upon a presentation I recently saw. They id'd the movement and shape of animals by that method (though the pca was used to identify shape). The classification of the movement information was used because different animals have different movement patterns. They had small films of a sonar where they traced the movment of the centroid over time (3-dimensional space because no info on depth). I think you could approach your problem in a similar way. This will be not the most sophisticated solution but it will probably give a vaiable solution.
Jul
7
comment What is the simplest way to classify airplan manuvers?
I would try something like functional pca or use a smooth version of the trajectory shape and build some clasification model upon it. Here I would suggest to try different methods to see how they perform.
Jul
2
awarded Curious
Jun
5
comment nlme() convergence error code = 1
How many different values of yield for each combination of break and state? As your error-code shows, nlme is using the nlminb-procedure. I've experienced strange behaviour of that procedure which for instance shows values for the target function for estimates which can in fact not be evaluated. I would recommend using optim if no bounds are specified and to reduce the number of break and state combination say in half, run nlme, if no problem occurs then increase until error occurs. Then look at the data. This trail and error ususally finds the error quite quickly.
Jun
5
comment nlme() convergence error code = 1
How many values of break do you have within each state, i.e., the cross table of break and state w.r.t yield?
Apr
24
asked What do I do with zero columns in a tensor.prod.model.matrix?
Apr
11
comment Setting up a naiv tensor product B-spline example
The solution is, that the hole thing, i.e., the tensor product, is defined row-wise. This means that $\boldsymbol{B}_{x,i}\otimes\boldsymbol{B}_{z,i}$ will always have row-dimension $1$ and only the column-dimension will expand.
Apr
9
revised Setting up a naiv tensor product B-spline example
deleted 11 characters in body; added 4 characters in body
Apr
9
revised Setting up a naiv tensor product B-spline example
deleted 11 characters in body
Apr
9
asked Setting up a naiv tensor product B-spline example
Mar
25
awarded Popular Question
1 2 3 4 5