khliland / emsc Goto Github PK

I was a little bit confused with QR decomposition in the code:

1. First, QR decomposition is done only for polynomial components (here). However, the original paper that is provided in resources suggest to do that for the whole model matrix
2. Then mldivide is used for solving the LS problem. And mldivide uses QR decomposition for that. Does it look like the QR decomposition is applied twice? Is there a reason for that?

In line 189 an absolute value is taken, but I did not found any explanation for that. Could you, please, explain it?

By looking at the code I could not understand what are constituents. From the code it looks like they are used exactly the same as interferents. For example, here they are excluded.
Another question is what is the difference between constituent and reference?

Here is a small example:

> x <- matrix(runif(50), ncol = 10)
> colnames(x) <- seq(420,600,length.out=10)
> x
           420       440        460       480        500       520        540       560        580       600
[1,] 0.1902495 0.6638507 0.96988143 0.4503034 0.02323665 0.7320788 0.44513545 0.5099727 0.00612236 0.9927394
...

> EMSC_model(x)
...
$abcissas
 [1]  1  2  3  4  5  6  7  8  9 10

I think the problem is in the line 144. Because is.ordered is actually related to factor variables, i.e. it always returns false for colnames.

Hello,

First of all, thank you for your package.
I was going though your code and I found that the code assumes that there is only one reference spectrum (for example, this line of code), however EMSC is not supposed to be limited to that. I would be good to allow using more than one reference spectra.

Hi,

I have an issue when using custom weights for the EMSC algorithm. Depending on the degree chosen, I get either "Longer object length is not a multiple of shorter object length" or dims [product xxxx] do not match the length of object [xxxx]"

The issue seems to be at line 15 :

modelw <- model$model * rep(model$weights, each = n)

I am not sure why the "n" variable is needed here, it doesn't match with the dimensions of the object "model". I know of the objectives of using EMSC but I am less familiar with the maths behind it.

Best regards,
Lucas