Closed FranciscoRMendes closed 3 years ago
\alpha
is not used in batch versions, but is in online algs. E.g., in another implementation from the class
package, note the lack of \alpha in in the batchSOM() function (https://www.rdocumentation.org/packages/class/versions/7.3-18/topics/batchSOM). But by default, \alpha
ranges for the kohonen
implementation as well as the SOM()
implementation from class
(e.g., https://www.rdocumentation.org/packages/class/versions/7.3-18/topics/SOM). So if this is included, even when calling batch
, it won't mess up computation (or help for that matter).
I was trying to change the $\alpha$ parameter i.e. the learning rate for the SOM based on this code chunk shared in class :
som_fit <- som(anes_scaled, grid = search_grid, alpha = c(0.1, 0.001), radius = 1, rlen = 500, dist.fcts = "euclidean", mode = "batch")
However, it seems that the documentation says that $alpha$ is not used for the batch algorithm. When I change the alpha rate the convergence does not speed up or slow down. When I did a Google search for the updating formula in Kohonen's 92 paper I see that : This equation does not have $\alpha$ Can you please tell me whether the hyperparameter alpha is used in the batch algorithm?