Hemos visto esta fórmula ya en varias ocasiones:
X = T.Pt+E
Solo algunos comentarios.
P(k.p) es la matriz de loadings, o de los (p más importanteseigenvectors) de la matriz de covarianzas S(k.k) de X(i.k).
T(i.p) es la matriz de scores.
“Para multiplicar matrices (A.B), el nº de columnas de la matriz A, tiene que ser igual al nº de filas de la matriz B, dando como resultado una matriz con el nº de filas de la primera y nº de columnas de la segunda”.
Tenemos que hacer la transpuesta de P, por tanto, para hacer el producto:
T(i.p).Pt(p.k) obteniendo como resultado una X(i.k) predicha, que al restarla de la original nos dará la matriz residual E(i,k).
Los datos entre paréntesis son(nº filas.nº columnas)
i=nº muestras
k=nº long. onda
p=nº eigenvalues seleccionados.
We have seen this formula on several occasions:
X = T.Pt+E
just a few comments:
P(k.p) is the matrix of loadings, or (more importants eigenvectors ) of the covariance matrix S(k.k) of X(i.k).
T(i.p) is the matrix of scores.
"To multiply matrices (A.B), the number of columns of the matrix A, must be equal to the number of rows of the matrix B, resulting a matrix with the number of rows of the first (A) and number of columns of the second (B)".
We have to do the transpose of P, therefore, to make the product:
T(i.p).Pt(p.k) resulting a predicted X(i.k) . If we substract this X predicted from the original, we
get the residual matrix E(i,k).
Data in parentheses are: (raw number, columns number)
i = number of samples
k = n ° long. wave
p = number of selected eigenvalues.
We have seen this formula on several occasions:
X = T.Pt+E
just a few comments:
P(k.p) is the matrix of loadings, or (more importants eigenvectors ) of the covariance matrix S(k.k) of X(i.k).
T(i.p) is the matrix of scores.
"To multiply matrices (A.B), the number of columns of the matrix A, must be equal to the number of rows of the matrix B, resulting a matrix with the number of rows of the first (A) and number of columns of the second (B)".
We have to do the transpose of P, therefore, to make the product:
T(i.p).Pt(p.k) resulting a predicted X(i.k) . If we substract this X predicted from the original, we
get the residual matrix E(i,k).
Data in parentheses are: (raw number, columns number)
i = number of samples
k = n ° long. wave
p = number of selected eigenvalues.
No hay comentarios:
Publicar un comentario