From what I can tell, there's no option to specify a covariance (instead of a corr matrix) in PCA() in package FactoMineR.
Any comments?
PCA(dd2_pcdat, scale.unit=FALSE)
Related
I need to get the principal coefficients (daily) of the first EOF of the daily geopotential height field (data = geopot.nc). I use only the cold season (N,D,J,F,M,A). My code is as follows:
cdo selmon,11,12,1,2,3,4 geopot.nc geopotw.nc
cdo --reduce_dim - copy geopotw.nc geopotw3.nc #delete 1 dimension of level=1)
cdo sub geopotw3.nc -ydaymean geopotw3.nc deseason.nc
cdo mulcoslat deseason.nc deseasoncos.nc
cdo eof,3 deseasoncos.nc eval.nc eof.nc
I got a warning message: statistics-module (Warning): Eigenvalue computation with one-sided jacobi scheme did not converge properly. 7017371 of 7244721 pairs of columns did not achieve requested orthogonality of 1e-12.
What I am doing wrong? Maybe I am missing something? Or maybe there are other better ways to achieve the result I am looking for? For example in R. Thank you for the help!
I'm not sure if this post will help, you can change the max iteration number using an environmental variable in the following way:
export MAX_JACOBI_ITER=100
See if that will help with your convergence issue?
The example of ML.NET for Matrix Factorization here has no evaluation method for testing the model. It's an one-class matrix factorization example, because its trainer option is set to options.LossFunction = MatrixFactorizationTrainer.LossFunctionType.SquareLossOneClass;
I searched further and found another matrix factorization example. Although it's not the one-class matrixfactorization but this time it has an evaluation. However the evaluation method is for a regression model. var metrics = mlcontext.Regression.Evaluate(prediction, labelColumnName: "Label", scoreColumnName: "Score");
Isn't matrix factorization a class of binary classification? Is it valid to use regression evaluation for the model? I actually put up this question a few times on the community chat for ML.NET but I've got no answers so far. Any comments or answers would be appreciated.
The problem that the second example is trying to solve is recommendation, if you look at line 55.
The metrics used for evaluation are the same as the regression metrics, that's why they are being used.
For more examples on Matrix Factorization you can look at the documentation:
https://learn.microsoft.com/en-us/dotnet/api/microsoft.ml.trainers.matrixfactorizationtrainer?view=ml-dotnet-preview
https://github.com/dotnet/machinelearning/blob/master/docs/samples/Microsoft.ML.Samples/Dynamic/Trainers/Recommendation
How can I obtain the coefficients of the regression function in the LMT leave nodes?
Thanks!
It should come up by default.
The screenshot below contains the coefficients that are generated using LMT:
This result was achieved without changes to LMT Default Parameters on a Randomly generated dataset and using Weka 3.7.11.
From Java, the LMT.ToString() method should give you the leaves of the tree.
Using a JSL script, I would like to extract the covariance matrix of a nonlinear model.
I have a 4PL curve. But when I request:
m["Logistic 4P"]["Parameter Estimates"]["Covariance of Estimates"]["Reference"][""];
It is said that it is an outlinebox and therefore can't be converted into a Data Table, nor a matrix.
However, while right clicking on it, I can convert it in both, so it must be possible using JSL.
Any ideas?
Ok finaly found it, If it can help someone:
m["Logistic 4P"]["Parameter Estimates"]["Covariance of Estimates"]["Reference"][1] << make into data table;
I'm trying to find a way to calculate EER value using Explorer in Weka but without sucess.
I can just see the ROC curve (thereshold curve).
Anyone can show how to find EER value using Weka.
Thanks.
I implemented a Weka package to calculate the EER.
The link is:https://sourceforge.net/projects/eer/?source=directory
Best Regards,